Patent Grant
11936734
Aspects described herein generally relate to computers, networking, hardware, and software. More specifically, some aspects described herein relate to a distributed and persistent computer-based simulation, including expressing query-based interest for entities and/or components of the simulation.
Conventional simulation systems are unable to scale to support very large numbers of objects to simulate those objects in real-time. Such systems have typically relied on a single instance of a simulation engine, running on a single physical or virtual computer system, to simulate the entire simulated world. Consumers of these simulation systems have had to choose between correctness, graphical fidelity, and real-time-interaction, with no solution offering the ability for all three on a large scale system. The magnitude and complexity of the situation is further increased if the consumer desires to simulate complex real-world problems which may require more computing power than a single simulation engine can provide. For example, a simulation of a city may require simulation of a large number of vehicles, pedestrians, bicyclists, traffic patterns, traffic lights, subway systems, transit vehicles, airplanes, and a multitude of other entities that affect and contribute to city life.
In one known approach, computing resources have been statically assigned to a portion of the simulated world. A disadvantage of this approach may be that as the simulated objects, actors, etc. move across the simulated world as the simulation progresses, the simulated objects may congregate on a very small region of the simulated world. If sufficient objects move to the very small region, the computing resources may be overloaded (resulting in slower processing), the simulation may terminate unexpectedly, and/or simulation data may be lost. Another disadvantage of this approach may be that state information of the simulation for a region may be concentrated on a single computing resource and may not be shared or spread across several resources, making fault tolerance or recovery from an unexpected termination difficult and time-consuming. In addition, this approach may not lend itself to easily support stateful migration of simulated objects across region boundaries, and thus simulations usually limit stateful migrations to only players.
These and other problems are addressed herein.
The following presents a simplified summary of various aspects described herein. This summary is not an extensive overview, and is not intended to identify key or critical elements or to delineate the scope of the claims. The following summary merely presents some concepts in a simplified form as an introductory prelude to the more detailed description provided below.
To overcome limitations in the prior art described above, and to overcome other limitations that will be apparent upon reading and understanding the present specification, aspects described herein are directed towards systems, methods, apparatuses, and computer-readable media storing computer executable instructions that, when executed, cause a system to perform a simulation by simulating a plurality of entities on a plurality of workers. Each entity of the plurality of entities may comprise one or more components. The system may run, on one or more machines, a plurality of bridges, and the plurality of bridges may be configured to facilitate data communications between the plurality of workers and one or more entity databases. Each worker of the plurality of workers may be assigned to a different bridge of the plurality of bridges. The system may modify an entity of the plurality of entities to include an interest component. The interest component may indicate a query subscription to the one or more entity databases and/or a frequency for receiving, from the one or more entity databases, updates for the query subscription.
In some examples, modifying the entity to include the interest component may be performed at a time corresponding to a time that the entity is created. Additionally or alternatively, the system may perform the simulation by receiving, during runtime of the simulation, a request to update one or more of the query subscription or the frequency for receiving updates for the query subscription. Based on receiving the request, the interest component may be updated to indicate one or more of an updated query subscription or an updated frequency.
In some examples, the system may perform the simulation by causing a worker of the plurality of workers to be authoritative over the interest component of the entity of the plurality of entities. Based on an entity-component authority of the worker and the interest component, an interest set of the worker may be modified to include the query subscription to the one or more entity databases and the frequency for receiving updates for the query subscription. Additionally or alternatively, the system may perform the simulation by causing a second worker of the plurality of workers to be authoritative over a different component of the entity of the plurality of entities. The simulation may be performed by subscribing, by a bridge of the plurality of bridges and assigned to the second worker authoritative over the different component of the entity, to the query subscription at the frequency for receiving updates for the query subscription. The bridge may send, to the second worker authoritative over the different component of the entity and at the frequency for receiving updates, updates to the one or more entity databases and corresponding to the query subscription.
In some examples, the interest component may be modified to indicate an updated query subscription to the one or more entity databases. Additionally or alternatively, the frequency may comprise a first frequency, and the interest component may indicate a second query subscription to the one or more entity databases, and/or a second frequency for receiving, from the one or more entity databases, updates for the second query subscription, wherein the second frequency is lower than the first frequency.
In some examples, the interest component may indicate a spatial element based on one or more of a position of the entity of the plurality of entities or a position of a different entity of the plurality of entities. Additionally or alternatively, the interest component may be based on one or more of a property of another component associated with the entity of the plurality of entities or a property of a component associated with a different entity of the plurality of entities. Additionally or alternatively, the frequency may comprise a first frequency, and the interest component may indicate a first spatial element associated with the first frequency, a second query subscription to the one or more entity databases, a second frequency for receiving, from the one or more entity databases, updates for the second query subscription, and/or a second spatial element associated with the second frequency.
A more complete understanding of aspects described herein and the advantages thereof may be acquired by referring to the following description in consideration of the accompanying drawings, in which like reference numbers indicate like features, and wherein:
In the following description of the various embodiments, reference is made to the accompanying drawings identified above and which form a part hereof, and in which is shown by way of illustration various embodiments in which aspects described herein may be practiced. It is to be understood that other embodiments may be utilized and structural and functional modifications may be made without departing from the scope described herein. Various aspects are capable of other embodiments and of being practiced or being carried out in various different ways. Additionally, the phraseology and terminology used herein are for the purpose of description and should not be regarded as limiting. Rather, the phrases and terms used herein are to be given their broadest interpretation and meaning.
As will be appreciated by one of skill in the art upon reading the following disclosure, various aspects described herein may be embodied as a method, a computer system, or a computer program product. Accordingly, those aspects may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, such aspects may take the form of a computer program product stored by one or more computer-readable storage media having computer-readable program code, or instructions, embodied in or on the storage media. Any suitable computer-readable storage media may be utilized, including hard disks, CD-ROMs, optical storage devices, magnetic storage devices, and/or any combination thereof. Particular data structures may be used to more effectively implement one or more aspects described herein, and such data structures are contemplated within the scope of computer executable instructions and computer-usable data described herein. In addition, various signals representing data or events as described herein may be transferred between a source and a destination in the form of electromagnetic waves traveling through signal-conducting media such as metal wires, optical fibers, and/or wireless transmission media (e.g., air and/or space.)
As a general introduction to the subject matter described in more detail below, aspects described herein are directed towards systems, methods, and techniques for providing a distributed and persistent simulation development environment. In some examples, the simulation development environment may also be spatially-optimized. In some other examples, the simulation development environment might not be spatially-optimized. Other aspects described herein may allow for the integration of existing non-distributed simulation programs into a large-scale distributed simulation. Yet other aspects described herein may be used to automatically and spatially balance and distribute the simulation workload.
Computer software, hardware, and networks may be utilized in a variety of different system environments, including standalone, networked, virtualized, and/or cloud-based environments, among others.
The various network nodes 123, 125, 127, 129, and 131 may be interconnected via a network 121, such as the Internet. Other networks may also or alternatively be used, including private intranets, corporate networks, local area networks (LAN), wide area networks (WAN), metropolitan area networks (MAN), wireless networks, personal networks (PAN), and the like. Network 121 is for illustration purposes and may be replaced with fewer or additional computer networks. Network 121 may have one or more of any known network topology and may use one or more of a variety of different protocols, such as Ethernet. Devices 123, 125, 127, 129, 131, and other devices (not shown) may be connected to one or more of the networks via twisted pair wires, coaxial cable, fiber optics, radio waves, or other communication media.
It will be appreciated that the network connections shown are illustrative and other means of establishing a communications link between the computers may be used. The existence of any of various network protocols such as TCP/IP, Ethernet, FTP, HTTP and the like, and of various wireless communication technologies such as GSM, CDMA, Wi-Fi, and WiMAX, is presumed, and the various computing devices in simulation system components described herein may be configured to communicate using any of these network protocols or technologies.
The term “network” as used herein and depicted in the drawings refers not only to systems in which remote computing devices are coupled together via one or more communication paths, but also to stand-alone devices that may be coupled, from time to time, to such systems that have storage capability. Consequently, the term “network” includes not only a “physical network” but also a “content network,” which is comprised of the data which resides across all physical networks.
The Input/Output (I/O) module 109 may include a microphone, keypad, touch screen, game controller, joystick, and/or stylus through which a user of the simulation computing device 101 may provide input, and may also include one or more of a speaker for providing audio output and a video display device for providing textual, audiovisual and/or graphical output. Software may be stored within memory 111 and/or storage to provide instructions to processor 103 for enabling a simulation computing device 101 to perform various actions. For example, memory 111 may store software used by a simulation computing device 101, such as an operating system 113, application programs 115, and an associated internal database 117. The database 117 may include a second database (e.g., as a separate table, report, etc.) That is, the information may be stored in a single database, or separated into different logical, virtual, or physical databases, depending on system design. The various hardware memory units in memory 111 may include volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. Simulation computing device 101 and/or computing devices 127, 129, 131 may also be mobile terminals (e.g., mobile phones, smartphones, personal digital assistants (PDAs), notebooks, etc.) including various other components, such as a battery, speaker, and antennas (not shown.)
Aspects described herein may also be operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of other computing systems, environments, and/or configurations that may be suitable for use with aspects described herein include, but are not limited to, personal computers, server computers, hand-held or laptop devices, vehicle-based computing devices, multiprocessor systems, microprocessor-based systems, programmable consumer electronics, network personal computers (PCs), minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like.
The one or more instances of the simulation runtime 248 within a virtual server 202 may communicate with each other to determine an instance which may serve as a master. For example, the simulation runtime 248 instances may utilize a consensus protocol to determine a master. A master simulation runtime 248 instance may be responsible for routing communications between the other simulation runtime 248 instances within the virtual server 202 and other simulation runtimes 248 executing in other virtual servers 202. As will be explained in greater detail below, the simulation runtime 248 may allow for distributed simulations where simulation workload is automatically distributed across available virtual server(s) 202. The virtual server 202 illustrated in
The virtual server 202 may comprise a hardware layer 210 with one or more hardware elements that communicate with the virtual server 202. Optionally, the hardware layer 210 may comprise one or more physical disks 212, one or more physical devices 214, one more physical processors 216, and one or more physical memories 218. Physical components 212, 214, 216, and 218 may include, for example, any of the components described above with respect to simulation computing device 101. In one example, physical devices 214 may include a network interface card, a video card, a keyboard, a mouse, an input device, a monitor, a display device, speakers, an optical drive, a storage device, a universal serial bus connection, a printer, a scanner, a network element (e.g., router, firewall, network address translator, load balancer, virtual private network (VPN) gateway, Dynamic Host Configuration Protocol (DHCP) router, etc.), or any device connected to or communicating with virtualization server 301. Physical memory 218 may include any type of memory. In another example, physical memory 218 may store data, and may store one or more programs, or set of executable instructions. Programs or executable instructions stored in the physical memory 218 may be executed by the one or more processors 216 of virtual server 202. Virtual server 202 may further comprise a host operating system 220 which may be stored in a memory element in the physical memory 218 and may be executed by one or more of the physical processors 216.
Hypervisor 230 may provide virtual resources to operating systems 246a-246n or to workers 249 executing on virtual machines 240 in any manner that simulates the operating systems 246 or workers 249 having direct access to system resources. System resources may include, but are not limited to, physical disks 212, physical devices 214, physical processors 216, physical memory 218, and any other component included in hardware layer 210. Hypervisor 230 may be used to emulate virtual hardware, partition physical hardware, virtualize physical hardware, and/or execute virtual machines that provide computing resources to simulation runtime 248 and workers 249. Hypervisor 230 may control processor scheduling and memory partitioning for a virtual machine 240 executing on virtual server 202.
Hypervisor 230 may be Type 2 hypervisor, where the hypervisor may execute within a host operating system 220 executing on the virtual server 202. Virtual machines 240 may then execute at a level above the hypervisor 230. The Type 2 hypervisor may execute within the context of a host operating system 220 such that the Type 2 hypervisor interacts with the host operating system 220. One or more virtual server 202 in a simulation system 200 may instead include a Type 1 hypervisor (not shown.) A Type 1 hypervisor may execute on a virtual server 202 by directly accessing the hardware and resources within the hardware layer 210. That is, while a Type 2 hypervisor 230 may access system resources through a host operating system 220, as shown, a Type 1 hypervisor may directly access all system resources without the host operating system 220. A Type 1 hypervisor 230 may execute directly on one or more physical processors 316 of virtual server 202, and may include program data stored in the physical memory 318.
The simulation runtime 248 may cause the hypervisor 230 to create one or more virtual machines 240 in which additional simulation runtime 248 and worker 249 instances may execute within guest operating systems 246. Hypervisor 230 may load a virtual machine image to create a virtual machine 240. The hypervisor 230 may execute a guest operating system 246 within virtual machine 240. Virtual machine 240 may execute guest operating system 246.
In addition to creating virtual machines 240, hypervisor 230 may control the execution of at least one virtual machine 240. Hypervisor 230 may present at least one virtual machine 240 with an abstraction of at least one hardware resource provided by the virtual server 202 (e.g., any hardware resource available within the hardware layer 210.) Hypervisor 230 may control the manner in which virtual machines 240 may access physical processors 216 available in virtual server 202. Controlling access to physical processors 216 may include determining whether a virtual machine 240 should have access to a processor 216, and how physical processor capabilities are presented to the virtual machine 240.
As shown in
Each virtual machine 240 may include a virtual disk 242a-242n (generally 242) and a virtual processor 244a-244n (generally 244.) The virtual disk 242 may be a virtualized view of one or more physical disks 212 of the virtual server 202, or may be a portion of one or more physical disks 212 of the virtual server 202. The virtualized view of the physical disks 212 may be generated, provided, and managed by the hypervisor 230. Hypervisor 230 may provide each virtual machine 240 with a unique view of the physical disks 212. Thus, the particular virtual disk 242 included in each virtual machine 240 may be unique when compared with the other virtual disks 240.
A virtual machine 240a-240n may execute, using a virtual processor 244a-244n, one or more workers 249a-249n using a guest operating system 246a-246n. The guest operating system 246 may be any one of the following non-exhaustive list of operating systems: WINDOWS, UNIX, LINUX, iOS, ANDROID, SYMBIAN. Guest operating system 246 may be a purpose-built operating system based on one or more of the aforementioned operating systems. For example, guest operating system 246 may consist of a purpose-built version of LINUX which may comprise only the functional modules necessary to support operation of the workers 249. Optionally, and as described in further detail below, a virtual machine 240a-240n may execute one or more bridge modules (not shown) corresponding to the one or more workers 249a-249n executing in the virtual machine 240a-240n.
Referring to
The simulation runtime 248 contains the program code to implement the elements and components which comprise the simulation environment, as described in further detail herein. For example, the simulation runtime 248 may comprise implementation code for one or more of the bridge modules of the cloud-based computing platform 310, as further described herein and as illustratively shown in
The cloud-based computing platform 310 may comprise private and/or public hardware and software resources and components. For example, a cloud may be configured as a private cloud to be used by one or more particular customers or client computing devices 340 and/or over a private network. Public clouds or hybrid public-private clouds may be used by other customers over open or hybrid networks. Known cloud systems may alternatively be used, e.g., MICROSOFT AZURE (Microsoft Corporation of Redmond, Wash.), AMAZON EC2 (Amazon.com Inc. of Seattle, Wash.), GOOGLE COMPUTE ENGINE (Google Inc. of Mountain View, Calif.), or others.
The simulation development environment 300 may be deployed as a Platform-as-a-Service (PaaS) cloud-based computing service which may provide a platform for allowing a user to develop, run, and manage a simulation. This may allow a user or client to create a simulation without understanding the intricacies of distributed computation or requiring access to infrastructure teams or supercomputers. The simulation development environment 300 may be delivered as a public cloud service from a provider. In such a scenario, client organizations may provide pre-existing models, simulations, and/or databases which may be integrated with the simulation development environment 300. Alternatively, the simulation development environment may be delivered as a private service within a private network of a client organization.
The cloud-based computing platform 310 may comprise one or more virtual servers 202a-202f (generally 202) such as the virtual server 202 illustrated in
The cloud-based computing platform system 300 may also comprise a cloud-based data store 320. The storage resources in the cloud-based data store 320 may include storage disks (e.g., solid state drives (SSDs), magnetic hard disks, etc.) and other storage devices. Alternatively, the cloud-based data store 320 may be provided by a known cloud-based storage provider, such as, AMAZON S3 (Amazon.com Inc. of Seattle, Wash.), GOOGLE CLOUD STORAGE (Google Inc. of Mountain View, Calif.), or others. Optionally, the cloud-based data store 320 may be implemented or deployed separately from cloud-based computing platform 310 as shown in
The cloud-based data store 320 may comprise one or more application assemblies 322. An application assembly 322 may comprise data which may define entities and components of a simulation, as well as, procedures which may define one or more behaviors of each of the entities and components in a simulation. Optionally, an application assembly 322 may comprise schemas, data structures, serialized objects, and the like which may define the entities and components which make up a simulation. Optionally, an application assembly 322 may comprise computer-readable code or instructions, scripts, statically-linked libraries, dynamically-linked libraries, and the like which may define one or more behaviors for the elements in the simulation. Virtual servers 202 in the cloud-based computing platform 310 may load an application assembly from the cloud-based data store 320. The simulation runtime 248 in each virtual server 202 may use the data and procedures comprised in an application assembly 322 to cause the execution of a distributed, persistent, and spatially-optimized simulation. The cloud-based data store 320 may also comprise initialization data and/or procedures 324 which define a starting or initial condition for a simulation. For example, the cloud-based computing platform 310 may load initialization data 324 from the cloud-based data store 320 which may cause a predetermined number of entities and components to be instantiated and initialized to a predetermined initial state. In another example, the cloud-based computing platform 310 may load and may execute one or more initialization procedures 324 which may cause a predetermined number of entities and components to be instantiated and initialized to a predetermined state. In yet another example, the entities and the components may be instantiated and initialized to a predetermined state based on a combination of initialization data 324 and initialization procedures 324 loaded by the cloud-based computing platform 310 from the cloud-based data store 320.
The cloud-based data store 320 may comprise a snapshot 326 of a simulation. A simulation snapshot 326 may define a valid state of a simulation, and may comprise data and/or procedures which may return a simulation to that valid state if or when it is loaded and/or executed by the cloud-based computing platform 310 from the cloud-based data store 320. The valid simulation state defined by snapshot 326 may be a known state or a desired state of the simulation. Optionally, the simulation state defined by snapshot 326 may be a previously saved state of a running simulation. A snapshot 326 may store some state of a simulation, which might not be a complete representation of a simulation at a certain time. When such a snapshot is loaded by the cloud-based computing platform 310, the user code within the workers may be able to derive a valid state.
A portion of the cloud-based computing platform 310 may be related, for example, one or more virtual servers 202 may be executing a simulation on behalf of the same end user, or on behalf of different users affiliated with the same company or organization. In other examples, certain virtual servers 202 may be unrelated, such as users affiliated with different companies or organizations. For unrelated clients, information on the virtual servers 202 or cloud-based data store 320 of any one user may be hidden from other users.
In some instances, client computing devices 340 may implement, incorporate, and/or otherwise include one or more aspects of computing device 101 and computing device 202. Client computing devices 340 may be any type of computing device capable of receiving and processing input via one or more user interfaces, providing output via one or more user interfaces and communicating input, output, and/or other information to and/or from one or more other computing devices. For example, client computing devices 340 may be desktop computers, laptop computers, tablet computers, smart phones, or the like. In addition, and as illustrated in greater detail below, any and/or all of client computing devices 340 may, in some instances, be special-purpose computing devices configured to perform specific functions.
The client computing devices 340 may comprise a worker integration library 342 and an instance of a worker process 249. A client computing device 340 may utilize the worker integration library 342 and the worker process 249 to connect to a simulation executing in the cloud-based computing platform 310. As described in further detail below, a client computing device 340 may receive data from the cloud-based computing platform 310 describing relevant portions of the simulation. The worker process 249 executing in the client computing device 340 may utilize that received data to render the relevant portions of the simulation on a display or other user interface device. The client computing device 340 may also transmit data and commands to cloud-based computing platform 310 which may affect the state of the simulation. The data and commands may be transmitted in response to user input. Optionally, the transmitted data and commands may be generated in response to calculations performed by the worker integration library 342 or the worker process 249.
Advantageously, and as illustrated in greater detail above, a simulation developer using a simulation development environment may be able to scale up a game or simulation to be considerably larger than would be possible using a single machine. In addition, the simulation development environment may allow for an arbitrary number of user participants and data sources to integrate into the simulation. Furthermore, the simulation development environment may remove the need for a simulation developer to worry about scalability or data synchronization among different parts of the simulation.
The state and behavior of an entity (e.g., 420, 430, and 440) may be determined by the combination of components (e.g., 421, 422, 423, 431, 432, 433, and 441) comprised by the entity. Each component (e.g., 421, 422, 423, 431, 432, 433, and 441) may comprise a subset of the state and behavior attributed to the entity (e.g., 420, 430, and 440) as a whole. For example, as shown in
Entities (e.g., 420, 430, and 440) may comprise properties which may be common across all entities. For example, entities (e.g., 420, 430, and 440) may comprise an identifier value which may be used to uniquely identify each entity instance within simulated world 410. Entities (e.g., 420, 430, and 440) may comprise properties which may be shared across multiple components. For example, entities (e.g., 420, 430, and 440) in a video game simulation may comprise position and velocity values since it is likely that most components in such a simulation may require access to those values. Additionally, locating commonly used properties within an entity may reduce coupling between the components and facilitate communication between the components of an entity.
Referring to
Components may comprise one or more properties 530. The state of a component 510 may be defined by the values held by the properties 530 comprised by the component 510. Similarly, the state of an entity may be defined by the values held by the properties 530 of all the components comprised by the entity. The value of a property may be, for example, a pointer or a URL to other data, e.g., a large asset. The state of a component 510 may be stored in local memory (e.g., 242a-242n, 244a-244n, 218) for access during execution of the simulation. Optionally, the state of a component 510 may be stored in cloud-based data store 320 as part of a snapshot 326 and thus may be persisted across simulation runs. The state of a component 510 may be stored periodically (e.g., continuously.) The rate at which the state of a component 510 is persisted may vary based on one or more factors. For example, if or when the state of a component 510 changes rapidly, the storage rate may also increase commensurate with the rate of change. In another example, the storage rate may be higher for properties which may require a higher degree of accuracy than other properties.
Where it is described that an entity or component may exhibit a certain behavior, it is to be understood that another element, such as a worker module, for example, may perform the required calculations on behalf of that entity or component and emit or receive the corresponding signals or data.
Events 550 may indicate the occurrence of a transient action on component 510. Component 510 may emit one or more events 550 in response to making a determination (or events 550 may be emitted for one or more components 510), reaching a particular result, receiving user input, or another type of trigger. Other components within the simulation may monitor the occurrence of an event 550 and update their state or perform an action in response to the event 550. The other components may be comprised by the same entity (e.g., a worker module) as the emitting component or may be comprised by other entities within the simulation. For example, a traffic signal entity in a city simulation may emit an event if or when the traffic signal indicator changes to red. A vehicle entity in the city emulation may receive the event and may come to a stop in response to the event. In another example, a rigid body component may emit an event if or when it has determined that it has collided with another object.
Optionally, procedures 540 may be used to update the values of properties 530, as well as, cause the component 510 to emit events 550. Procedures 540 may also receive and process commands 520 from other components and/or the simulation runtime 248. Thus, procedures 540 may define the behavior of component 510 within the simulation. Alternatively, a simulation runtime 248 may delegate to a worker 560 the implementation of the behavior of component 510. In such a scenario, simulation runtime 248 may delegate write access of properties 530 and events 550 from component 510 to worker 560. Component 510 may have at most one writer assigned to it at any one time. A worker may make component updates and trigger events for a component it is authoritative for. Any worker can send a command to a component belonging to a specific entity, and that command may be handled at the worker, which is authoritative for that component. Optionally, a worker 560 may implement the behavior of a component based on real-time and/or real-world behavior of a physical entity being simulated. For example, a worker 560 may periodically collect position, velocity, and direction data from one or more sensors mounted on a vehicle or other moving object and use that information to modify properties 530 and emit events 550 of component 510. In another example, a worker 560 may receive previously recorded real-world position, velocity, and direction data of a vehicle or other moving object and use that information to modify properties 530 and emit events 550 of component 510. Thus, a worker 560 may be used to incorporate real-time and/or real-world into the simulation. Any other real world objects, people, events, and/or systems may be used to generate data as input for a simulation.
Delegation may require specification of a worker constraint which may identify a type of worker capable of simulating the behavior of component 510. Worker 560 may be one of a plurality of worker types which may be specialized to perform certain kinds of computations. Workers 560 may only understand a subset of the components (e.g., 421, 422, 423, 431, 432, 433, and 441) that define entities (e.g., 420, 430, and 440) within a simulation 410. For example, in a city simulation, one worker type may simulate vehicle positions, another worker type may simulate traffic signals, and yet another type may simulate environmental emissions.
Worker 560 may comprise data structures and/or objects and software programs to simulate the behavior of a subset of the components (e.g., 421, 422, 423, 431, 432, 433, and 441) within a simulation 410. Worker 560 may be a process corresponding to one or more aspects of workers 249, as described in
In order to simulate the behavior of a component (e.g., 421, 422, 423, 431, 432, 433, and 441), worker 560 may need information (e.g., properties, events) from nearby entities (e.g., 420, 430, and 440) within simulation 410. For example, a worker simulating a traffic intersection in a city simulation may need information from vehicles in nearby intersections, but not from vehicles which are miles away from the intersection. The interest region for worker 560 may comprise all regions comprising nearby entities (e.g., 420, 430, and 440) from which the worker 560 needs information. The interest region for worker 560 may comprise entities (e.g., 420, 430, and 440) for which worker 560 is not authoritative. The simulation 410 may automatically synchronize the data between worker 560 and the other workers which are authoritative for the nearby entities.
Worker 560 may communicate with the simulation 410 (e.g. with entities) via a bridge 610, as illustrated in
Communication between bridge 610 and worker 560 may be effectuated via a worker application programming interface (API). Optionally, worker 560 may include user code and potentially other frameworks that the code runs within. The user code may use a worker API 630 to interact with the runtime, such as via the bridge. Worker API may allow a worker 560 which may have been developed independently from the simulation development environment to possibly function within and be managed by bridge 610. Optionally, the worker API may allow for the integration of pre-existing non-distributed simulation programs into a large-scale distributed simulation. For example, a game engine (e.g., UNITY by Unity Technologies SF of San Francisco, Calif.) may be integrated into a simulation to simulate rigid-body physics or to provide client-side rendering and navigation. In another example, a multi-modal traffic flow simulation software package (e.g., open source MATSIM, or other commercially available software packages) may be integrated into a city simulation. Other worker engines or programs may alternatively or also be used.
In another example implementation, worker 560 may require special-purpose hardware or other physical resources that might not be available within a cloud-based platform 310. In such a scenario, the worker API 630 and bridge 610 may reside on a computing device physically located remotely from the cloud-based platform 310 and may connect to the cloud-based platform 310 via the Internet or another type of network. Such a worker 560, which may reside outside of the cloud-based platform 310, (e.g., may execute on client devices 340a-340n) may be referred to as an external worker. And another worker 560, which may execute within the cloud-based platform 310, (e.g., may execute on servers 240a-240n, 202a-202f) may be referred to as an internal worker. Any one or more of the features described with reference to the cloud-based platform 310 may be used in or with this example implementation.
The worker API may allow a bridge to add or remove components of entities from the interest region of a worker (e.g., from the worker's view of the world), notify a worker of component state changes, delegate a component to a worker or to remove the delegation, signal component state changes for components on which the worker is authoritative, among other related functionality as described herein. In some examples, components of an entity's canonical representation may also be dynamically added to or removed from an entity database.
Among the functions provided by the worker API may be functions for adding or removing an entity. Optionally, worker API 630 may comprise a handler method to be called by bridge 610 when an entity enters the interest region of worker 560. For example, Method 1 is one example of a method signature that may be used to add an entity to the interest region of worker 560.
Method 1:
void OnEntityAdd(EntityId eid, EntityState initialState); where
eid is a value which may uniquely identify the entity being added; and
initialState is a data structure and/or object which may describe the initial state of the entity being added.
Although Method 1 is provided as an example for adding an entity to the interest region of worker 560, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. Method 1 may then be passed to a RegisterEntityAddHandler( ) worker API function, which may cause the Method 1 handler to be called whenever an entity should be added.
Optionally, worker API 630 may comprise a handler method to be called by bridge 610 when an entity leaves the interest region of worker 560. For example, Method 2 is one example of a method signature that may be used to remove an entity from the interest region of worker 560.
Method 2:
void OnEntityRemove (EntityId eid); Where
eid is a value which may uniquely identify the entity being removed.
Although Method 2 is provided as an example for removing an entity from the interest region of worker 560, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. Method 2 may then be passed to a RegisterEntityRemoveHandler( ) worker API function, which may cause the Method 2 handler to be called whenever an entity should be removed.
The worker API may also comprise functions for notifying a worker that the properties of a component within the worker's interest region have changed state. For example, worker API 630 may comprise a handler method to be called by bridge 610 when the properties of a component within the interest region of worker 560 have changed state. Method 3 is one example of a method signature that may be used to notify worker 560 of the changed state.
Method 3:
void OnStateChanged_Component1(EntityId eid, SomeState state); where
eid is a value which may uniquely identify the entity which may comprise the component whose properties changed state; and
state is a data structure and/or object which may describe the state of the component.
Although Method 3 is provided as an example for notifying worker 560 of a changed state, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. In some variants, the state parameter may comprise only the subset of properties of the component that have changed since the last update, for efficiency. Method 3 may then be passed to a AddComponentStateChangeHandler( ) worker API function, which may cause the Method 3 handler to be called whenever the properties of a component within the worker's interest region have changed state.
Among the functions provided by the worker API may be functions for dynamically changing component authority assignments. Worker API 630 may comprise a handler method to be called by bridge 610 when worker 560 may now be authoritative for a component. For example, Method 4 is one example of a method signature that may be used to delegate component authority to worker 560.
Method 4:
void OnComponentDelegate(EntityId eid, ComponentId cid); Where
eid is a value which may uniquely identify the entity which may comprise the component being delegated; and
cid is a value which may uniquely identify the component being delegated.
Although Method 4 is provided as an example for delegating component authority to worker 560, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. Method 4 may then be passed to a RegisterComponentDelegateHandler( ) worker API function, which may cause the Method 4 handler to be called whenever worker 560 may now be authoritative for a component.
Optionally, worker API 630 may comprise a handler method to be called by bridge 610 when worker 560 may no longer be authoritative for a component. For example, Method 5 is one example of a method signature that may be used to remove delegation authority for a component from worker 560.
Method 5:
void OnComponentUndelegate(EntityId eid, ComponentId cid); Where
eid is a value which may uniquely identify the entity which may comprise the component being undelegated; and
cid is a value which may uniquely identify the component being undelegated.
Although Method 5 is provided as an example for removing delegation authority for a component from worker 560, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. Method 5 may then be passed to a RegisterComponentUndelegateHandler( ) worker API function, which may cause the Method 5 handler to be called whenever worker 560 may no longer be authoritative for a component. In a distributed system, callbacks may be called sometime after the bridge has sent the corresponding messages. For example, this could mean that a worker may believe itself to be authoritative for a component when it no longer is, and vice-versa.
In yet other examples, worker API 630 may comprise a handler method to be called by bridge 610 for setting or unsetting a worker 560 as authoritative for a component. For example, Method 7 is one example of a method signature that may be used to set or remove delegation authority for a component for worker 560.
Method 6:
void SetIsAuthoritative(EntityId eid, ComponentId cid, Boolean isAuthoritative); where
eid is a value which may uniquely identify the entity which may comprise the component;
cid is a value which may uniquely identify the component; and
isAuthoritative is a true/false value which may indicate whether to set or unset worker 560 as authoritative for a component.
Although Method 6 is provided as an example for setting or unsetting a worker 560 as authoritative for a component, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure.
Changes to a database, such as an entity database, may be made by any process, such as a worker, some other external system, or the runtime system itself (e.g., load balancing enforcers, as will be described in further detail below). For example, a process may make changes to a canonical state in the database in response to some other change(s) in the canonical state. As another example, the entity database may have some internal systems that may update the canonical state, potentially based on some other state change. User code within a worker may be notified, via worker APIs, of the changes made to components that the worker is not authoritative for. Worker API 630 may comprise a method to be called by worker API 630 when the properties of a component for which worker 560 is authoritative have changed state. Method 7 is one example of a method signature that may be used to update the properties of the components for which worker 560 is authoritative.
Method 7:
void UpdateState_Component1(EntityId eid, SomeState state); Where
eid is a value which may uniquely identify the entity which may comprise the component whose properties changed state; and
state is a data structure and/or object which may describe the updated state of the component.
Although Method 7 is provided as an example for updating the properties of the components for which worker 560 is authoritative, various other methods and/or functions may be used. For instance, other parameters may be included in the method without departing from the disclosure. Method 7 may be called whenever the properties of a component for which worker 560 is authoritative have changed state.
Optionally, worker 560 may be configured to periodically send a heartbeat signal to bridge 610. If or when worker 560 ceases to transmit heartbeat signals, bridge 610 may determine that worker process 560 may have terminated unexpectedly. In response to the determination, a higher-level system, such as a worker scheduler (as will be described in further detail below), may terminate the bridge 610 and request that a replacement worker process 560 (and new counterpart bridge 610) be allocated and instantiated.
Examples of Dynamic Load-Balancing of Data Subscriptions via Hierarchical Aggregators and Connection Migrations
In a simulation, such as a computer game, the simulation may have some form of state, parts of which are to be available to participants in the simulation (e.g., players of a game). The state may continually change, and the changes may be shown to one or more participants. There may be many participants, states, and frequent changes to states. In an online game, for example, there may be many players, a lot of game state, and/or frequent changes to states. These changes may be caused by the players themselves and/or by internal systems, such as by an artificial intelligence (AI) system controlling the behavior of non-player characters (e.g., a monster entity, a tree entity, and the like).
The game simulation 900 may comprise the distributed game server 906. The distributed game server 906 may be configured to provide simulation functionality for the players 902A-D and other players, non-player entities (e.g., monster entities, weapon entities, tree entities, etc.), and other portions of a game world. As previously explained, the game server 906 may comprise one or more physical servers, one or more virtual servers, or a combination thereof. The game server 906 may comprise distributed game states 908, which may represent the state of entities, such as player and non-player entities, and/or components for the entities. The game server 906 may also comprise game systems 910, such as AI for non-player entities, rigid-body physics for players and other entities, and other game systems. An efficient and scalable data distribution mechanism may be used to accommodate a large number of players, a lot of game state, and/or frequent changes to game state. Other types of simulations with many participants, states, and/or state changes (e.g., simulations of epidemics, city simulations, trading simulations, etc.) may similarly benefit from an efficient and scalable data distribution mechanism.
Simulation states and/or state changes may be made available to clients via a data source, which may be available on a single machine or a plurality of machines. The data that the data source stores and/or makes available to clients may be referred to as the data domain. Simulations may comprise pieces of data, and the data may be of different data types, such as scalar, spatial, proto, or other data types. Scalar data may comprise a scalar value, such as a single value, a double, or other scalar types. Spatial data may comprise a vector describing a point in space. For example, a spatial's x, y and z components may be given by s1, s2 and s3 respectively. Proto data may comprise a protocol buffer (protobuf) comprising fields, each of which may be a piece of data. Proto data may contain other proto data. An entity may be a piece of data of type proto. The fields of this proto may be the entity's states, and the subfields of those states may be properties. One of the entity's states may be a position or spatial location within the simulated world.
A query may refer to a particular subdomain in a data domain. For example, a query may comprise a predicate that matches some data items. If a query q matches some piece of data x, we write q(x), and if it does not we write ¬q(x). Various query types may be supported, including numeric ranges, bounding boxes, bounding spheres, projections, Boolean combinations, etc. Numeric ranges may be represented by the expression in range (lo, hi). This query may match a scalar if and only if (scalariff) it is within the given range: q(x)≡x∈[lo, hi]. lo and/or hi can be infinite, which may allow expression of a numeric range such as x≤10 as inrange(−\infty, 10). Bounding boxes may be represented by the expression bbox (bl, tr). This query may match a spatialiff it is within the bounding box with bottom left corner bl and top right corner tr: q(x)≡∀1≤i≤3. bli≤xi<tri. Bounding spheres may be represented by the expression distance(center, radius). This query may match a spatialiff it is within the bounding sphere with the given center and radius: q(x)≡∥x−centre ∥≤radius. Projections may be represented by project (field, q′). This query may match a protoiff the field of that proto matches the query q′: q(x)≡q′(x.field). For Boolean combinations, the simulation may take arbitrary Boolean combinations of queries, e.g., q1∧q2, q1∨q2, ¬q, TRUE, FALSE. An example query is InSphere(position: (40, 85, 48), radius: 20 m) AND HasComponent(Inventory). Another example query is InCone(position: (10, 50, 23), range: 100 m, fov: 100 deg) AND Health.current_health<50.
A query language may be used to express a sub-domain or query of a data domain, such as the data of a data source. The data source (e.g., an entity database, as will be described in further detail below) may support a query language to define queries that may refer to a subset of the data source's state. For example, a query may correspond to a view of the data source, and the view may comprise a subset of the data source's state. The data source may be optimized for certain types of queries, such as for queries that refer to a particular region of a simulated space.
An SQL-like language may be used to express queries. For example, the expression:
SELECT*FROM Entity WHERE states.position WITHIN(Vector3(0, 0, 0), Vector3(10, 10, 10))
may be used to match entities whose position is within a bounding box [(0,0,0), (10,10,10)].
A plurality of clients, which may be running on different machines, may connect to the data source (e.g., an entity database) via a network. The data source may support subscriptions. After connecting to the data source, a client may subscribe to one or more queries, such as by using a query language. A query may comprise a streaming query. For a streaming query, an update to the data source may result in an update being sent to a client if the update changes data corresponding to a streaming query subscribed to by the client. For example, after a client subscribes to a streaming query, the client may continually receive, from the data source, data that is within the domain of the streaming query. The client may receive data via a subscription if there is an update to the data. Additionally or alternatively, the client may periodically receive data via its subscription. The client may build up a view of the data source for the streaming query or queries the client is subscribed to. Because the data at the data source may be updated at any point, such as by the clients themselves or by some other means (e.g., other clients, internal systems, etc.), the clients may continually receive updates relevant for their streaming query subscriptions. The process may aim (e.g., continually) to synchronize the view at the client to the view at the data source corresponding to the streaming query or queries. A client may close a streaming query, so that updates for that streaming query are no longer sent back to the client. A client may also modify a streaming query and may receive database updates corresponding to the modified streaming query. In some scenarios, an update sent to a client might not be an update received by the client with a streaming query subscription. For example, the update may arrive at some point later in time, potentially merged with and/or overwritten (e.g., partially or completely overwritten) by one or more other updates that followed that update, potentially from different senders.
Clients may directly connect to a data source and subscribe to one or more queries offered by the data source.
In some examples, aggregators may act as a proxy between clients and a data source. Clients may connect to the data source via one or more of the aggregators and subscribe to one or more queries offered by the data source using the one or more aggregators. Sending data out may be a resource-intensive task, which may cause a bottleneck in the system. In order to reduce the amount of data sent out to clients by a data source, another layer of machines (e.g., aggregators) may be used to aggregate queries. For example, an aggregator may combine a client's queries and subscribe to a combined query that includes the queries of the clients connected to the aggregator, as will be described in further detail below.
Returning to
Overlap of clients' queries may be increased to reduce the amount of data being transmitted between clients and data source(s) and to conserve bandwidth. For example, the overlap of clients' queries connected to an aggregator can be increased by controlling which aggregator each client connects to. An aggregator controller, which may comprise a software module that may be located on a separate machine, may be used to modify the connections. For example, an aggregator controller may be aware of the number of aggregators, and the aggregator controller may logically split a data domain into a number of sub-domains the same as the number of aggregators. Accordingly, a mapping from the sub-domain (which can be expressed as a query) to an aggregator may be created. Instead of pre-splitting a data domain into a fixed number of sub-domains (e.g., of equal sizes), the aggregator controller may dynamically partition the data domain according to incoming client queries so that roughly the same number of clients are connected to each aggregator.
Assume, for example, that the aggregator controller 1312 determines that the client 1302 is to connect to the aggregator 1308A. After the aggregator controller 1312 determines the aggregator for the client 1302, the aggregator controller 1312 may send, to the client 1302 and/or the aggregator 1308A, a response indicating that the client 1302 is to connect to the aggregator 1308A. Based on the response, the client 1302 may connect to the aggregator 1308A, as shown in
In some examples, clients may be dynamically reconnected to different aggregators.
The aggregator controller 1412 may connect to and communicate with the aggregator 1408A, the aggregator 1408B, or other aggregators (not shown) to, for example, i) determine a good partitioning of their clients, and ii) to tell clients (e.g., indirectly) which aggregator to connect or switch to. The aggregator controller 1412 may also be connected to a data source 1404, depending on whether the aggregator controller 1412 can get information from the data source 1404 to execute its partitioning strategy.
The aggregator controller 1412 may dynamically redistribute clients among the aggregators in order to reduce traffic sent by the data source 1404. For example, queries may be mostly spatial. A query may be spatial when the query in one way or another restricts the simulation space of the data domain. For example, if a query contains InSphere(<center>, <radius>), it may restrict the data domain to that subset of the simulation space. If the queries are mostly spatial, then a possible algorithm for redistribution may (i) distribute the space among the available aggregators, and (ii) dynamically tell each client to connect to the aggregator that corresponds to a center point of the client's query or queries. For example, the client may connect to the aggregator that corresponds to the sub-domain query with which the client's query has the largest overlap. Because clients may dynamically change their queries, clients may be dynamically told to reconnect to a different aggregator.
As another example, the aggregator controller may logically partition the data domain among the aggregators according to the type of information. For example, if a client (e.g., a worker) connects and is interested in only component X of any entity, regardless of where this entity is in the simulation space, the aggregator controller may tell the client to connect to the aggregator responsible for component X. This way, if, for example, 100 such clients connect to that aggregator, the aggregator might only receive a single stream of updates for component X from its data source.
In step 1504, the aggregator controller 1412 may send, to the aggregator 1408A, a request for the client 1402B connected to the aggregator 1408A to migrate to a different aggregator 1408B. In step 1506, the aggregator 1408A may send, to the client 1402B an indication that the client 1402B will soon be connected to the aggregator 1408B. The indication may indicate the aggregator 1408B and/or indicate a time for the migration.
In step 1508, the client 1402B may begin temporarily storing (e.g., buffering) data to be sent. After the client 1402B receives the migration message (e.g., in step 1506), the client 1402B may stop sending updates to the aggregator 1408A and instead may start buffering the updates. In some examples, the client 1402B may immediately stop sending updates to the aggregator 1408A, e.g., on the order of a round trip time (RTT). The client 1402B may also tell the aggregator 1408A that it has started buffering updates. The client 1402B may enter a different mode of operation for the transition period, where updates are buffered instead of sent. The data that the client 1402B buffers may comprise data that the client 1402B is continually sending (e.g., updates to the data store 1404).
In step 1510, the aggregator 1408A may determine whether temporary storage of client 1402B's data has been completed. For example, the client 1402B may notify the aggregator 1408A when the client 1402B has completed its buffering process. The aggregator 1408A may also stop sending data, such as data associated with queries, to the client 1402B. If temporary storage at the client 1402B is not complete (step 1510: N), the aggregator 1408A may wait until temporary storage is complete. If temporary storage at the client 1402B is complete (step 1510: Y), the method may proceed to step 1512.
In step 1512, the aggregator 1408A (e.g., a source aggregator) may send, to the aggregator 1408B (e.g., a target aggregator), information indicating the current (e.g., latest) version or view of the data that the aggregator 1408A sent to the client 1402B or otherwise knows about for the subscription. For example, the aggregator 1408A may send, to the aggregator 1408B, information about the connection state with the client 1402B (e.g., information about which parts of the query view(s) have been synchronized). By sending this information, the need to resend some information (e.g., information that the client 1402B already has) may be reduced, saving network bandwidth for other uses. The aggregator 1408A may communicate this information directly with the aggregator 1408B, or may indirectly communicate with the aggregator 1408B, such as via the aggregator controller 1412. The aggregator 1408A may also send, to the client 1402B, a request for the client 1402B to connect to the new aggregator 1408B.
In step 1514, the client 1402B may disconnect from the aggregator 1408A and connect to the new aggregator 1408B. With brief reference to
As previously explained, the aggregator 1408B may receive (e.g., in step 1512) information indicating the current version of the data that the client 1402B has, such as its view. The aggregator 1408B might not need to resend the same data to the client 1402B now connected to the aggregator 1408B, resulting in less data being transmitted and more efficient usage of available network bandwidth. By performing one or more of the steps shown in
The aggregator controller may also adjust the queries based on the new connections. For example, queries q1 and q2 shown in
In some scenarios, combining aggregators may cause an aggregator to be overloaded. For example, the aggregator controller may know that combining aggregator 1608A and 1608B would cause the combined aggregator 1708A to be overloaded. The aggregator controller may track (e.g., through the help of the aggregators) how much data is going through the different connections that match the different queries involved, such as the connections between the aggregators and the data source and/or the connections between the aggregators and the clients. As will be described in further detail below, there may be multiple layers of aggregators (e.g., two layers, three layers, ten layers, etc.). The aggregator controller may also track how much data is going through the different connections between aggregators, which may be in different layers. Based on the amount of data going through one or more of the connections, the aggregator controller may determine a target configuration (e.g., the number of aggregators and the connections of the aggregators). For example, the aggregator controller may attempt to keep, for one or more of the connections, the amount of data going through the connection below a threshold amount of data. As will be described in further detail below, the aggregator controller(s) may also attempt to keep the total amount of data going through a plurality of connections (e.g., the connections for the configuration or the connections for a portion of the configuration) below a threshold amount of data. In some examples, the aggregator controller may receive an indication that the combined aggregator 1708A is overloaded after (e.g., immediately after) the combination. For example, the aggregator 1708A may now send, to clients C1-6, the same amount of data as both aggregators 1608A and 1608B together before. The system may introduce another level of aggregator(s) to alleviate the load of the overloaded aggregator.
As described above, the aggregator controller may change the configuration of aggregators from the configuration shown in
In general, extra layer(s) in an aggregator hierarchy may be introduced as necessary and/or to improve the efficiency of the system by removing overload scenarios. An aggregator controller may be connected to the aggregators (e.g., all of the aggregators) and/or the data store. The aggregator controller may determine (i) an overall layout of aggregators, (ii) whether to add and/or remove any aggregators, and/or (iii) whether to change what a client or an intermediate aggregator is connected to. The intermediate aggregator may use the same or a similar reconnection protocol as described for clients above.
A plurality of aggregator controllers may be used, such as if there is too much work for a single aggregator controller. Each of the plurality of aggregator controllers may handle a portion of the aggregator hierarchy.
In step 1904, the queries or subdomain(s) that clients subscribe to may be determined (e.g., by a developer or a master aggregator controller). Clients may choose their own queries, and the types of queries that clients choose may depend on particular use cases. For example and with reference to
Returning to
As previously explained, in addition to analyzing the amount of query overlap, the aggregator controller may analyze how much traffic matches each sub-part of the queries involved. How much queries overlap may be an indication of how much less data could be sent by combining the queries. For example, if two queries overlap, but no data going through the connection matches the conjunction or overlap of the two queries, there might not be a savings. If, on the other hand, all or a substantial amount of data going through matches the overlap, combining the queries may result in traffic savings, even when the overlap is small. As explained above, the aggregator controller(s) may attempt to reduce the total amount of data going through a plurality of connections (e.g., the connections for a particular configuration of aggregators or the connections for a portion of the configuration).
In general, the system, such as via one or more aggregator controllers, may consider one or more different factors to determine a configuration of aggregators and connections for the system. The aggregator controller(s) may attempt to cause correct delivery of updates, from senders to data source(s) and/or from data source(s) to end-clients. For example, the aggregator controller(s) may attempt to avoid overloading data source(s) and/or any intermediate aggregators. The aggregator controller(s) may do this by reducing or minimizing the amount of traffic being sent from a data source or an aggregator to its clients. This, in turn, may be done by dynamically analyzing and exploiting the patterns of data being sent and moving clients' connections in order to get the benefit from clients' query overlaps for the data being sent. In other words, the system may attempt to minimize the amount of duplicate information being sent in order to achieve scalability. As previously explained, the amount of query overlap and the amount of data being transmitted through each connection may be used to determine the configuration to reduce the transmission of duplicate information. An estimation of how much less or more data would be sent by combining or splitting queries and how overloaded or underloaded the corresponding parts of the system are may determine whether to introduce or remove a node (e.g., an aggregator) from the system or combine nodes in the system. The aggregator controller(s) may also attempt to use fewer machines (e.g., as few machines as possible) to reduce or minimize costs.
With brief reference to
The aggregator controller may evaluate the traffic corresponding to the sub-queries making up the combined query, such as at an instance in time or over a period of time. The aggregator controller may determine the amount of transmitted data corresponding to overlapping portions of the combined streaming query. For example, the aggregator controller may determine the amount of transmitted traffic corresponding to the overlap of query a and query b, the amount of transmitted traffic corresponding to the overlap of query a and query c, the amount of transmitted traffic corresponding to the overlap of query b and query c, and/or the amount of transmitted traffic corresponding to the overlap of query a, query b, and query c.
The aggregator controller may additionally or alternatively determine the amount of transmitted data corresponding to the non-overlapping portions of the combined streaming query. For example, the aggregator controller may determine the amount of transmitted data corresponding to query a (but not query b or query c), the amount of transmitted data corresponding to query b (but not query a or query c), and/or the amount of transmitted data corresponding to query c (but not query a or query b). Assume that, based on the determination of the amount of transmitted data, the aggregator controller determines that in practice, while the overlap of the streaming query for client C3 between clients C1 and C2 is favorable, the data throughput corresponding to the overlapping portions of the query c with query a and query b has minimal overlap. For example, the distribution of updates to queries might not be uniformly distributed across queries. The aggregator controller may determine to remove client C3 from the group, and the aggregator with clients C1 and C2 connected may subscribe to a different streaming query, such as query d (820). For client C3, the aggregator controller may calculate heuristics for other potential groups and may assign client C3 to a different group it deems favorable or may assign client C3 to its own aggregator.
As previously explained, the amount of query overlap and/or the amount of transmitted data may be used to determine the proper configuration of aggregators and connections. One or more other factors may be used to determine the configuration. For example, the aggregator controller may consider resource usage on the aggregator. If an aggregator is overloaded or running near capacity (e.g., memory and/or CPU usage), the aggregator controller might not add further queries or clients to the aggregator. The aggregator controller may also consider constraints on certain connections from client to aggregator, aggregator to aggregator, or aggregator to data source. For example, the aggregator controller may consider whether a connection has a maximum threshold bandwidth, and the aggregator controller might not add another query that would cause the amount of transmitted data to exceed the threshold bandwidth. The aggregator controller may also consider the geographic distribution of served clients. For example, if an aggregator is serving several clients, it may be desirable for the purposes of consistent latency to assign clients from the same geographic region to the same aggregator or groups of aggregators.
Returning to
The system may also generate one or more other aggregators, which may be at a different layer. For example, the system may generate an aggregator 1808A shown in
Returning to
As previously explained, a query aggregation may comprise a widening of a set of queries. For example, given a set of queries Q={q1, q2, . . . }, a widening ΔQ (also written as, for example, Δiqi) may be determined. A widening of a set of queries may itself be a query that satisfies:
∀x·∀q∈Q·q(x)⇒(ΔQ)(x)
In other words, for each piece of data x, if there is some query in Q that matches x, then ΔQ also matches x. For example, ΔQ may overapproximate the set Q. A consequence may be:
¬(ΔQ)(x)⇒¬∃q∈Q·q(x)
For example, if some data x does not match ΔQ then no q∈Q matches x either. A widening can match many more data items than the original set of queries did. In particular, TRUE may be a valid widening for any set of queries.
For example, assume the existence of queries Q={inrange(0,10), inrange(15, 20)}. Some possible widenings of Q may be:
ΔQ=inrange(0,10)∨inrange(15,20)
ΔQ=inrange(0,20)
ΔQ=inrange(0,∞)
ΔQ=TRUE
Widening for query and data indices may also be defined. For a query index, it may be the widening of the set of queries in the index, and for a data index D:
∀x·x∈D⇒(ΔD)(x)
As before, ΔD may overapproximate D. If some data item x does not match ΔD then x∉D. Similarly, ΔD may be as precise as desired. For example, ΔD may match exactly the items in D; ΔD may match everything; or ΔD may be anything in between.
For example, assume the existence of the data index D={0,4,10}. Some possible widenings of D may be:
(ΔD)(x)≡x=0∨x=4∨x=10
ΔD=inrange(0,10)
ΔD=TRUE
An entity cache may contain some subset of the entities in the world. A cache may have local processes that write directly to the data in the cache. Caches may be connected together into a network. Pairs of caches may share some subset of entities. Writes can occur at any cache in the network. Caches may exchange messages in order to synchronize data between themselves. For example, a cache C1 may send a message to another cache C2, instructing the cache C2 to update its view of the state of the world. These messages may comprise commands. As another example, a local process may add an entity with ID 1337 to cache C1. Cache C1 may send an AddEntity(id=1337) message to C2, instructing the cache C2 to add the entity to its view of the world.
A cache may also emit events describing mutations that have occurred to the cache. For example, a local process may update the position of entity 1337 from (0,0,0) to (0,0,1) in cache C1. Cache C1 may emit the event, e.g.:
StateUpdated(id: 1337, old_state: {position: (0, 0, 0)}, new_state: {position: (0, 0, 1)}, update: {entity_id: 1337, position: (0, 0, 0)}
The events emitted by a cache may comprise the cache's event stream.
Networks of caches may be synchronized.
Recursive queries may be used. For a cache Ci to synchronize its clients Cij with other caches in the system, cache Ci may subscribe to its peers with a sufficiently large query that each event happening elsewhere in the system may be routed to Ci and hence to its interested clients. For a cache Ci with clients Cij, the cache's query may be given by:
The server Ci may subscribe to the TRUE query on each of its clients so that events originating on a client (e.g., in the tree rooted at Ci) make their way to Ci. Server Ci may also construct an up to date of the union of the data in its tree.
Synchronized caches may be implemented. A cache may contain some set of entities, and the cache may know about some set of other caches. Each of these caches may be subscribed to a particular query. Local processes may search the data of the entities in the cache and may mutate the data of entities in the cache. This may be implemented efficiently by using a data index to store the entities and a query index to store the queries for each connected cache. If a client cache updates its query, the server cache may recalculate its widened query and (if it has changed), update its own query amongst its peers. For example, the client's updated query might include data not within the previous widened query. When a client or peer updates its query from q to q′, the system may determine whether there are new entities to add or remove. To find the entities to add, the query q′∧¬q may be executed against the data index. To find the entities to be removed, the query ¬q′∧q may be executed.
A cache or local process may subscribe to a cache's event stream (e.g., a streaming query). As previously described, the subscriber may subscribe to a query on the subscribee. The query may be matched against each emitted event, and matching events may be translated into commands that the subscribee may send to the subscriber. The commands may describe the changes the subscriber is to apply in order to reflect the state described by the event. For example, a cache C1 may want to know about position updates happening to spiders in the bounding box [(0,0,0), (10,10,10)] at the cache C2. The cache C1 may subscribe to the following query:
SELECT update.position FROM SyncEvent WHERE new_state.position WITHIN((0, 0, 0), (10, 10, 10)) AND new_state.is_spider=TRUE
As another example, a local process may want to be sent a message whenever a light switch is turned on while a cable is attached, or when a cable is attached to an on switch. It may subscribe to the query:
SELECT update FROM SyncEvent WHERE NOT(old_state.switch=TRUE AND old_state.cable=TRUE) AND (new_state.switch=TRUE AND new_state.cable=TRUE)
As yet another example, a cache C1 may want to receive about 10% of the position updates from entities between 1 km and 5 km away from (0, 0, 0). It may subscribe to the query:
SELECT update FROM SyncEvent WHERE INRANGE(new_state.position, (0, 0, 0), 5000) AND NOT(INRANGE(new_state.position, (0, 0, 0), 1000)) AND RAND( )<0.1)
Examples of Architecture(s) Used for Query-Based Simulations
The worker layer 705 may comprise a plurality of workers. The workers may comprise managed workers, such as managed workers 730a-730c, and unmanaged workers, such as client workers 720a-720c. Managed workers may be managed by the platform, such as for the managed worker's lifetime. Managed workers may run on machines, such as virtual machines. The simulation runtime may decide which managed workers run on which virtual machines. The simulation runtime may also move a worker from one virtual machine to another, such as by stopping one and starting another. An unmanaged worker may be controlled externally, such as for the unmanaged worker's lifetime. For example, an unmanaged worker may comprise an external worker that an end-user of an application runs to connect to a simulated world. Each client machine may comprise one client component, or may comprise multiple client machines. Managed and unmanaged workers may connect to the platform.
Client workers 720 and managed workers 730 may incorporate and/or otherwise include one or more aspects of worker 560 as illustrated in
Through a worker's connection or otherwise, the worker may act as a client, such as a client of a database. For example, the worker may subscribe to zero or more database subscriptions or queries in order to update its local view or views of the world. In addition, the worker may request updates to the database via its connection. Changing the number and/or sizes of machines, such as virtual machines, in each layer may be achieved by, for example, moving components from one machine to another. For example, managed workers may be moved by killing one and starting an equivalent.
The bridge layer 715 may comprise a plurality of bridges. The bridge layer 715 may act as a distributed gateway, such as the distributed gateway 904 shown in
Each bridge may hold, for example, two views of the world: one view of the entities that its worker is interested in seeing as the entities exist in the database layer 725, and one view of the entities as they are seen by the bridge's worker. The bridge may be responsible for making the two views consistent with each other, figuring out which side needs to be told what for the views to agree on the state of the world. A worker view may comprise the bridge's knowledge about what the worker has already been told about (e.g., for the query or queries it is subscribed to). A database view may comprise the bridge's knowledge about what the worker view is to be. For example, the bridge may receive, from the database, an update to the database view. The bridge may accept the database view, store the database view, and/or compare the database view to the worker view. If the worker view differs from the database view, the bridge may send, to the worker, an update with the new or updated values from the database view. The bridge may also update its view of what the worker sees (e.g., the worker view) to correspond to the database view.
If the worker sends an update to its bridge, the bridge may send the update to the database for updating the canonical state of the database. For example, the update from the worker may bypass the database view and the worker view and be sent directly to the database. If the update is successfully applied at the database, the database may propagate its updated view back to the bridge, such as if the subscribed query matches the update sent. The bridge may receive the update to the database view and perform one or more of the steps previously described. For example, the bridge may store the updated database view, compare the database view to the worker view, send an update to the worker and/or update the worker view if the worker view differs from the updated database view, etc.
The bridge layer 715 may comprise a plurality of database clients, such as database client 750a, database client 750b, or other database clients. The database clients may comprise software components that connect bridges to one, some, or all database shards in the database layer 725. As shown in
Some of the components shown in
Changing the number and/or sizes of machines, such as virtual machines, in each layer may be achieved by, for example, moving components from one machine to another. For example, bridges may be modified via bridge migration. During a simulation, the runtime may perform a bridge migration. For example, the runtime may tell a worker to dynamically connect to a different bridge, which may be on a different bridge service. As the simulation 700 progresses, bridge 740a may be designated to be migrated from machine 710d to machine 710h. In such a scenario, a new bridge instance (not shown) may be instantiated in machine 710h, and client worker 720a may be temporarily connected to both bridge 740a and the new bridge instance while the bridge migration is effectuated. Once the migration is completed, client worker 720a may be disconnected from bridge 740a, and bridge 740a may be terminated. Alternatively or additionally, bridge 740a may be terminated in machine 710d and restored on machine 710h. For example the bridge may be restored from data temporarily stored within the runtime, or from data stored in the database (e.g., entity database). Snapshots, such as snapshot 326, may generally be used for taking a backup copy of the entity database (e.g., the whole entity database) on a semi-regular basis, e.g., every 10 minutes or some other interval.
As previously explained, the runtime simulation may put bridges of workers, which may have overlapping subscription queries, on the same bridge service. Accordingly, the runtime may reduce traffic from database shards in the database layer 725 to bridge services 710d and 710h by reducing duplication. This may be done by aggregating queries into a single (e.g., larger) query. When an aggregator (e.g., a bridge service holding multiple bridges) receives data from one of its data sources (e.g., a database shard through the database bridge) matching its query, the aggregator may forward this data to its clients of which queries also match the received data.
Returning to
The database layer 725 may comprise, for example, a distributed, in-memory database for a simulation. For example, the database may comprise an entity database, a portion or all of the states of a simulation's entities may be stored in the entity database. The entity database may be persisted (e.g., regularly) as snapshots to disk or another medium for longer-term storage. The entity database may also be restored from a snapshot. The entity database may also store the platform's internal information. Various examples of entity databases are described in U.S. application Ser. No. 16/042,068, filed Jul. 23, 2018, and entitled “ENTITY DATABASE,” which is herein incorporated by reference in its entirety.
The database layer 725 may comprise a plurality of machines 710j, 710k, etc., which may comprise virtual machines. Each virtual machine may comprise one or more components running within the virtual machine. For example, the virtual machine 710j may comprise a plurality of database shards, such as database shard 770a, database shard 770b, or other database shards. The virtual machine 710k may comprise a database shard 770c and/or other database shards. Each database shard may hold data corresponding to a distinct subset of the data domain. This may be achieved by splitting the data domain by entity ID (e.g., modulo the number of database shards). Each database shard may then act as a root data source, such as the data sources described above (e.g., data source 1004, 1104, 1304, 1404, or 1604). Thus while the system of
Each virtual machine may comprise a database bridge, which may comprise a connection endpoint for the database clients (e.g., database clients 750a, 750b, etc.) on one side and database shards (e.g., database shards 770a-c) on the other. Clients of the database may connect to some or all entity database shards in order to, for example, make reads or writes to the database. For example, the database bridge 760a may connect the database client 750a to the database shard 770a and/or the database shard 770b. Similarly, the database bridge 760a may connect the database client 750b to the database shard 770a and/or the database shard 770b. The database bridge 760b may connect the database client 750a to the database shard 770c. Similarly, the database bridge 760b may connect the database client 750b to the database shard 770c. While the database clients 750a and 750b may each be connected to database bridge 760a and database bridge 760b, the database clients 750a and 750b might not receive all messages from both bridges. For example, the database clients 750a and 750b may receive messages comprising updates to queries subscribed to by the corresponding workers.
Changing the number and/or sizes of virtual machines in each layer may be achieved by moving components from one virtual machine to another. For example, the number and/or sizes of database shards may be changed by merging or splitting the shards (and the data that the shard is responsible for). The database may be sharded by, for example, entity ID. The database may also support updates to its state. These updates may be atomic across multiple components of the same entity or even across multiple entities, depending on database sharding strategy.
Examples of Distributable and Customizable Load-Balancing
As previously explained, a simulation (e.g., a simulated world) may comprise entities, and these entities may have one or more components. The components may comprise data fields that may be changed. The simulation platform described herein may allow for running of code (e.g., user code) in workers in order to, for example, modify the component data fields. A worker may have authority over an entity-component, such as having write access to that entity-component. An entity-component may be assigned to a worker or may be simulated by a worker when that worker has authority over the entity-component. The authority information may be stored as, for example, part of an entity's data. Additionally or alternatively, authority may be transient, and authority may be sent to workers as a message on change of authority.
The authority component may indicate for the entity 1234 which worker is authoritative over the other components of the entity 1234. For example, Worker B (e.g., worker 2130b shown in
Bridges may subscribe to queries from the database 2110, which may return entities the corresponding workers are authoritative over. For example, the bridge 2120b may subscribe to a query 2150 to return entities that the worker 2130b is authoritative over. An example query may be:
Subscription query (SELECT*FROM entities WHERE authority=‘Worker B’)
The database 2110 may return, for example, data indicating that the worker 2130b is authoritative over the position component of entity 1234, the mass component of the entity 1234, and other entity-components that worker 2130b is authoritative over.
Worker load-balancing may be used to determine which worker is to be responsible for each component on each entity and/or to balance loads so that there are the right number of workers to simulate a world correctly. In particular, worker load-balancing may comprise a suite of tools and/or strategies that may help with one or more of the following: (i) how to perform lifecycle management of workers (e.g., starting, stopping, restarting on crash, etc. of workers), (ii) how to give a worker authority over an entity-component to allow the worker to simulate it, (iii) how to determine which workers are to be authoritative for which entity-components, or (iv) how to determine the number and which types of worker(s) are to be running in a given deployment of the simulation platform.
Strategy intent enforcement may be performed per simulation layer (e.g., physics and chat), and the load-balancing strategy may be split into distinct parts, with each part managing a different layer of the simulation. For example, assume that a simulation has three layers: one for physics, one for simulating weather, and one for a chat system. These layers may be configured and load balanced independently of each other. For example, the layers might not share workers. This may allow for choosing different strategies for each layer. For example, the physics and weather workers may be set up to have the world divided up by physical space, but the chat system may be divided up by user (e.g., player) ID ranges instead. Because a strategy intent may be a representation of how a simulated world is to be load-balanced, it can be configured (dynamically) by developers on the simulation runtime. This may allow developers to write their own load-balancing strategies.
The system 2300 may comprise one or more enforcement services used to implement load-balancing in a simulation, such as a large virtual world simulation. An enforcement service may take a strategy intent and may apply it to entities within its domain. As previously explained, a domain may comprise a subset of the entities in the world and may be defined through a query (e.g., an entity database query). In some examples, each enforcement service may act on a unique domain that is different from the domain that other enforcement services act on. Using a single enforcement service to act on a unique domain may prevent the enforcement services from interfering with each other. Because the simulated world may dynamically change its size and computational requirements, domains may be modified dynamically.
A strategy intent may map intended workers to their partitions. A strategy intent may comprise data with, for example, two (or more) fields: an entity database query (e.g., an entity query constraint) and an intended worker (e.g., a virtual worker) for that query (e.g., worker that components satisfying the query are to be assigned to). An enforcement service may be used so that entities within its domain, which match a particular partition query, are delegated to the corresponding worker.
Intent may be separated from enforcement or action for assigning authority over entity-components to workers. For example, one part of the system may be responsible for intent (e.g., determining how many and which kinds of workers are used to simulate the world), while another part of the system may be responsible for enforcement (e.g., starting and stopping workers). The technique may be used for managing workers. For example, both unmanaged workers and managed workers may be assigned to virtual workers. The following table shows an example of intent and action for simulating entities and for having workers.
Intent may be implemented by load balancing decision-making, as previously explained. Action may be implemented by load balancing enforcement. The intended existence of a worker may be referred to as a virtual worker. A worker scheduler service may be responsible for starting, stopping, and/or restarting workers to make sure that the virtual workers have corresponding workers running. Because the strategy intent may specify the intended worker for each partition, the list of virtual workers for the simulation may be derived from the strategy intent.
With reference to
The system 2300 may comprise the partition enforcer 2310 or a plurality of partition enforcers, which may be spread across several servers. These partition enforcers may share servers with other services to reduce the number of virtual machines used to run a simulation. Alternatively, the partition enforcers may reside in their own servers, such as to isolate computing capacity. Each partition enforcer, of a plurality of partition enforcers in the simulation, may have a domain, which may be spatial, entity id range, or some other way of sharding out partition enforcement such that the partition enforcers do not interfere with each other. A partition enforcer 2310 may receive, from the load-balancing strategy algorithm 2332, the set of partitions 2333. Each partition may indicate a query that specifies a subset of entity-components in a simulation.
Partition authority logic in the partition enforcer 2310 may be used to implement the load-balancing strategy (e.g., as determined by the eye in the sky device 2330). Specific authority logic in the partition enforcer 2310, on the other hand, may be used if a component desires a specific authority assignment. Entities may carry a self-defining read/write permission that describes who is eligible to gain authority over their components. This may either describe a specific type of simulation (e.g., “I require simulation by something belonging to the physics layer”) or a specific connected worker (e.g., “I require simulation by the worker with id 1234”). In the latter case, the component in question might not be participating in the load balancing system because it requires a specific authority assignment (e.g., worker with id 1234). Accordingly, the assignment for this component might bypass going through partition authority logic in the partition enforcer 2310.
The partition enforcer 2310 may send, to the database 2315, authority updates (if any), based on the output of the load-balancing strategy algorithm 2332. Each partition enforcer 2310 may subscribe to a query or data domain corresponding to the partition or partitions the partition enforcer is responsible for. The partition enforcer 2310 may receive, from the database 2315, data corresponding to one or more domain subscriptions. For example, the query may be a streaming query, and the partition enforcer 2310 may receive updates each time there is an update to an entity that satisfies the streaming query. Alternatively, the partition enforcer 2310 may periodically query the database 2315 for updates to entities in its partition. A domain may define a query that the partition enforcer 2310 subscribes to. The entity database 2315 may fulfill that subscription with entity update data. The entity database 2315 may send data corresponding to the query subscribed to by the partition enforcer 2310. The subscriptions can be provided as an initial configuration, or dynamically configured via a service.
A partition enforcer may need to know about the partitions relevant to its domain. The partition enforcer may filter out partitions that are not relevant to its domain, so that the simulation may be more easily scaled. If a domain query does not overlap with a partition query, it may be filtered out. For example, if the domain of a partition enforcer is “entities with an odd entity id” and the partition is “entities with an even entity id,” the query may be filtered out. More sophisticated filters may be used for more sophisticated domains or partitions (e.g., regions of 3D space that do not overlap). As another example, for component and/or position based queries and where partition enforcers are responsible for regions of space, the enforcers may throw away any partitions that do not apply to the components they care about or to the region of space they care about. The partition enforcer may attempt to determine which entities in its domain have component assignments that do not match any partitions. For those entities with an invalid assignment, the partition enforce may set the intended worker component to have a matching partition. Different methods may be used to achieve this. For example, the partition enforcer may subscribe to an entity database query that returns the count of entities that are within the partition enforcer's domain and incorrectly assigned. The partition enforcer may react when this count is not zero. As another example, the partition enforcer may subscribe to all entities within its domain and make the partition validity checks within the enforcer.
The system 2300 may comprise the virtual process manager 2325 (e.g., a virtual worker manager). Similar to the eye in the sky device 2330, the virtual worker manager 2325 may be co-located with other services to save on the number of servers used or might not be co-located with other services. The virtual worker manager 2325 may receive, from the load-balancing strategy algorithm 2332, the set of virtual processes (e.g., workers) 2334. The set of virtual processes 2334 may comprise a list of virtual processes that are to exist, and the virtual process manager 2325 may determine whether the physical processes exist and start and stop processes as necessary. The virtual worker manager 2325 may manage a virtual to physical mapping between workers. For example, the virtual worker manager 2325 may communicate with the worker runner or layer 2320 to start and stop workers and create new workers as needed. The virtual worker manager 2325 may start and stop physical processes to run the virtual workers. The virtual worker manager 2325 may also receive, from the bridge server or layer 2318 (or one or more other services), information about the state of the worker, such as whether it is connected, whether it has crashed, whether the latency is low, whether there are resources to run it with, etc. For example, a physical process satisfying a virtual worker may crash during a simulation, and the virtual worker manager 2325 may start up another physical process to replace the crashed physical process. The virtual worker manager 2325 may also receive information indicating the number and types of workers connected to the bridges in the bridge layer 2318. The partition enforcer 2310 and the virtual worker manager 2325 may be used to implement the strategy intent determined by the load-balancing strategy algorithm 2332 and based on data received from the load-balancing strategy algorithm 2332.
In general, a simulated world may have a plurality of partitions, and each partition may correspond to a different metric that entities may satisfy. For example, partitions may be bounded regions, such as spheres, cylinders, boxes, rectangles, or other shapes.
Returning to
Partition components may be placed on different kinds of entities. For example, assume that the simulation comprises a plurality of managed physics workers, one managed auction house worker that is to be global, and a plurality of players. Each player entity (or each player's client's virtual worker entity) may have a partition component, which may be used so that the player gets authority over his or her controls. An example partition component on a player entity or player's client's virtual worker entity is as follows:
There may be a partition component on several load-balancing entities placed throughout the world to spatially distribute physics workers. An example partition component on a load-balancing entity is as follows:
There may be a partition component on a virtual worker entity for a singleton auction house worker. An example partition component on the auction house worker is as follows:
As explained above, a partition component may be placed on different kinds of entities, depending on the particular load-balancing case the system is dealing with. For unmanaged workers (e.g., players), the lifetime of the partition may correspond to the lifetime of the player. This example may assume that the player entity goes away when the player disconnects. However, if the player entity does not go away when the player disconnects, the partition may be placed on a worker entity associated with the player client. For spatially load-balanced workers (e.g., physics), activities may occur in terms of regions of space, so the load-balancing entities in the world may describe the local partitions. For singleton or global workers, the partition may be placed on the singleton's virtual worker entity itself.
Use of partition components may allow for worker load-balancing. This may also provide a way of baking coarse-grained load-balancing intent into a snapshot, which may allow the simulation to handle snapshots taken in the middle of load-balancing transitions more gracefully. The partition component expressing load-balancing may also allow building tooling around it (e.g. rendering partitions in an inspector). The partition component of load-balancing intent may also allow building internal runtime features more easily. For example, a worker manager may easily determine the workers that need starting in order to fulfill a partition set.
A subscription policy (e.g., checkout policy) of managed workers may be based on the region they are intended to be authoritative on, rather than the entities they are currently authoritative on. This may mean that the scenario of managed workers unnecessarily checking in and out entities may be avoided. For example, if a managed worker knows not only about the immediate entities it has ownership over to simulate, but potentially a larger query describing what it might simulate, the managed worker can pre-load assets or pre-populate caches so that simulating entities in the regions of the query (but not near the existing entities) may be easier.
In some scenarios, two or more partitions may overlap. An enforcement service may be used to decide which of the corresponding intended workers to assign. A conflict resolution approach that may be used may be to change worker assignment when the current assignment does not match any of the intended workers that qualify. This way, oscillation between different intended workers may be avoided, which gives hysteresis and may be good for performance reasons (e.g., since switching authority may have an associated cost). Each partition enforcer may be aware of partitions in the world, such as all of the partitions.
Many of the entity components may belong to a single partition, but some entity components may belong to a plurality of partitions, resulting in overlapping partitions. A desired behavior for overlapping partitions may be defined. For example, transitions of an entity component between different partitions may be determined.
Users may configure overlapping intended authority bands to prevent authority thrashing (e.g., spatial hysteresis), such as to address the scenario that multiple partitions are valid. If the intended behavior is instead for there to be only one worker authoritative for one region of space, then the configuration may be set up with no overlapping partitions.
Examples of Query-Based Interest in a Simulation
As previously explained, a simulated world may comprise a plurality of entities, and the state(s) of each entity may be stored in components on those entities. Workers may perform processing and can update the state of the components. In some scenarios, one worker may be designated for each entity to be authoritative over the entity's state. In systems with, for example, complex entity interactions, each worker may be able to observe some subset of the entities in the simulated world in order to decide how to update its authoritative entities. What a worker is to observe to simulate its authoritative entities may be the worker's interest, and the set of entities described by the worker's interest may comprise the worker's view of the world.
In some simulation systems, a paradigm of what each worker can see may be used. For example, workers may be assigned to a particular location in the simulated world. A worker's view of the world may be constructed based on that assigned location and, for example, shapes associated with that location (e.g., a sphere of radius 30 meters around the worker's position).
In some simulation systems, a paradigm of what each entity can see may be used. A simulation may express interest for entities and their components (e.g., instead of interest for workers). Interest may be defined for each entity, and the worker may inherit the interest of the entity because the worker may be simulating the entity. The interest may be spatial, but might also not be spatial. In some examples, the entity may comprise a plurality of components, including an interest component. For example, a player entity may comprise an interest component and one or more other components, such as a physics component, a health component, a chat component, and/or other components. An entity interest may comprise the things that are needed by a worker to simulate a given entity. For example, the entity interest of a player entity might be “entities within 20 meters of the player entity.” An entity view may comprise a set of entities which satisfy the entity interest at some point in time. Where entity-based interest is used, a worker view may comprise a union of the entity views of entities for which the worker is authoritative. Entity-based interest may provide more flexibility and scalability for a system with complex entity types because entities might not be treated equally regardless of needs.
When views are associated with entities, the simulation system may able to understand much more complex views. Examples of interest regions comprising circles and spheres were previously described. However, other basic or complex shapes (e.g., cylinders, rectangles, cubes, custom shapes, etc.) may be used to define interest regions. Additionally or alternatively, the simulation system may use a query language to define interest. Simple shape queries may be represented using a query language that includes simple shape queries, and the queries may be applied on an entity basis, resulting in a more flexible simulation system.
As a brief example of a query language to define a shape query, if the interest of an entity is a sphere of radius 20 around an entity, the interest may be represented as:
Other types of queries may be supported, and a query language may be expanded in various unique ways. For example, the query language and their queries may optionally include various elements, such as frequency, component value matching, and/or referential interest.
In some examples, the query language(s) to express interest may support one or more frequency elements, which may allow more complex views in some scenarios. The frequency of the worker receiving updates may be adjusted for one or a plurality of interest regions. Instead of all views representing the real-time status of a simulation, a worker may choose to receive updates less frequently if, for example, the components that worker is simulating are not time sensitive. By specifying the frequency that workers receive updates, less information may be sent through the simulation's network, resulting in bandwidth and network resources being freed up for other uses.
In some examples, the query language(s) to express interest may support component value matching. As previously explained, interest may be based on position. However, by expressing interest as a query, the query language may be expanded to additionally or alternatively specify interest based on particular component values. As an example, an entity can express an interest query of “all entities where the team component is equal to team_A,” and the authoritative worker may receive updates from entities having a team component of “team_A.” Component value matching may be combined with position and/or frequency elements, as previously described. For example, in a simulated combat environment between two teams, different techniques may be combined to provide a very rich world. A player entity could support, for example, the following three queries: (1) details about players within 20 meters with live updates, which may simulate the local visible environment, (2) position, gear, height, and/or weight of any players within 100 meters updated 6 times per second, which may simulate human sight at a distance, and (3) position of any player who is on the player entity's team 1 time per second, which may simulate GPS location synchronization with team members.
In some examples, the query language(s) may support referential interest (e.g., interest based on other entities). Because an interest query can be evaluated by the system which stores a simulated world's state, interest queries may support indicating interest in other entities. As previously explained, an entity may express interest in a radius of a sphere around itself (or another dimension of another shape around itself). However, an entity may also or alternatively define its interest to include a sphere or other shape around another entity. The entity may also or alternatively base its interest on another entity's position. As previously explained, queries need not be spatial, and the interest might not be based on any entity's position. The worker authoritative over the entity expressing interest based on a target entity might have authority over the target entity or might not have authority over the target entity. In either case, the worker may receive updates based on interest expressed relative to the target entity. Referential interest may be combined with position elements, frequency elements, and/or component value matching, as previously described. Expanding on the simulated combat environment described above, a player could also control a drone. The player entity's interest query could be supplemented with position, gear, height, and/or weight of any players within 10 meters of the drone entity, updated 3 times per second. This may simulate, for example, the limited camera technology of a controlled drone. Moreover, with referential interest, there may be a plurality of players having interest around the same drone.
Interest may be expressed on each entity's component separately, which may allow flexibility in defining interest regions. In some scenarios, each entity-component might only be authoritative on a single worker. For example, a worker may become authoritative over a single component of an entity (e.g., the simulation may happen at the entity-component granularity). As previously explained, a simulation runtime may divide entities into components, and each component may be authoritative on a single worker. However, different components on the same entity may be authoritative on different workers. By attaching an interest query to an entity-component (e.g., instead of the entity), an entity-component authority model may be applied to the interest.
Interest query on an entity-component may allow the simulation system to associate functionality to components. For example, assume that a game simulation supports a map. The data used to represent entities on the map may be placed on a map component. An interest query used to populate data for the map can query specifically for this map component. Then any entity can be added or removed from the map based on whether the entity has the map component. The map component is now more than just a collection of data. It may be a marker of whether an entity supports a type of functionality. Examples of interest expressed on entity-components will be described below, such as in a MapRepresentation component.
An interest component may comprise a mapping of a component identifier to a query and/or a frequency, and an interest component may be added as a component to one or more entities. The map component field may be updated with component updates sent from workers, and a developer may have the ability to update the interest queries at runtime. These queries may be written in an extension of a public-facing query language, and may then be converted to an internal query language used by an entity database. Various examples of entity databases are described in U.S. application Ser. No. 16/042,068, filed Jul. 23, 2018, and entitled “ENTITY DATABASE,” which is herein incorporated by reference in its entirety.
An example component field for an entity may encode a mapping of {component_id−>(query, frequency)} for that entity. The bridge may interpret this mapping as if the bridge is authoritative for “component_id” on this entity. This mapping might only be relevant to workers that are authoritative on at least one component on the corresponding entity. For other workers, this mapping might not have an effect, and the corresponding bridge might not assume authority based on this mapping. Authority assignment and enforcement may be performed by a load-balancing system, as previously described. The bridge may subscribe, with a specified frequency, to updates from a data source (e.g., an entity database) which satisfy the query. For example, the bridge may select an interest where a particular worker has authority over the interest. The bridge may update its query subscriptions based on queries in the interest component of the corresponding authoritative worker. In these examples, queries may comprise query-frequency pairs, such as for each simulated entity-component. Frequency may be specified for different queries. Moreover, different frequencies may be specified for different queries. A query in the interest component may be updated via component updates.
Default queries may be specified. For example, a default query may be specified per component_id in the bridge configuration. If there is no default query explicitly set in the interest component for an entity-component, the runtime may automatically set the interest query for the entity-component based on some preconfigured query. For example, users may be able to express, in the bridge configuration, a default interest and frequency for one or more component types (e.g., PlayerControls or other component types):
Also or alternatively, each entity may contain an interest component, such as:
uint32 may hold the component ID (e.g., the PlayerControls component from above may be assigned a component ID, and that component ID may be included in the uint32 field). InterestConfig may comprise a list of one or more queries. map<uint32, InterestConfig> may comprise a type of a field called component_interest. Accordingly, component_interest may comprise a map that maps component types (e.g., referred to by their unique IDs) to the corresponding queries (e.g., listed in InterestConfig, which may comprise one query or multiple queries). If InterestConfig lists a plurality of queries, the system may take one or more union of the plurality of queries to create a single query or fewer queries. The interest component may be added by a user at the time of entity creation, or later if the simulation system supports it. For example, if a map contains an entry for PlayerControls, the associated query-frequency pair may override a default setting (e.g., specified in the bridge configuration) for PlayerControls. If such a mapping is subsequently removed, the interest set may revert to the default interest. The interest component of an entity may also be updated at runtime, such as by an authoritative worker for the interest component. As previously explained, the components of an entity may be authoritative on multiple workers, but each entity-component might be authoritative on one worker. These updates may update, at runtime, the query and/or frequency of interest for the worker(s) authoritative for the corresponding entity-components. Because the authoritative worker for the interest component may make changes to the interest component, users may choose whether they want a client-side or server-side authoritative model for controlling changes to the interest set.
Managed workers may be used for, for example, managing authority over single entity component or managing interest in one place. A managed worker may update an interest component explicitly. Authority of an interest component may be held by a single managed worker. The managed worker may receive requests from other workers via commands or component updates that the managed worker observes. The managed worker may be one of the managed workers authoritative on the components of the entity, or may be a dedicated managed worker that just manages interest changes. Workers may be able to modify their interest based on which entity components the workers are currently simulating. Based on what a managed worker observes about a client and/or through the client making a request to change its interest (e.g., via a command), the managed worker may update the client's interest. As previously explained, the client's interest may be stored in a component on the entity, such as the interest component. Workers may modify the client's interest using similar component updates as what workers use for modifying other components on the entity. Thus the update mechanism may be unified, resulting in improved and more efficient usability. When a worker becomes authoritative over a component on a particular entity, the worker's interest set may be automatically updated (e.g., expanded) to include the query associated with that entity-component. For example, assume that a worker becomes authoritative over a PlayerControls component in a game simulation. The worker's interest set may be expanded to include, for example, the components transform, inventory, and/or health for entities within 100 meters at a frequency of 30 hertz.
Various examples of interest components will now be discussed. For example, assume the following components on a player entity:
In a local environment, a position component may create a spherical interest, such as:
In this example, the component_id of the position component may be 54, and the interest may result in returning other player entities (e.g., corresponding to an id 2000) within 20 meters. In addition to the entry for component ID 54, there may also be entries for component ID 2004, component ID 2005, and/or any other component IDs. The interest may be specified in an InterestComponent on the player entity, along with the other components on the player entity listed above.
In a distant environment, the position component_interest from above may be updated with:
ReturnComponents may be used to receive component states or updates. For example, ReturnComponents(2000) may be used to send PlayerInfo, which has a component ID of 2000. ReturnComponents(2000,2001) may be used to send PlayerInfo and Appearance. In the first distance example InSphere(20), if a player entity is over 20 meters away, the system might not return the appearance for the player entity because players might look like opaque figures at that distance. The interest may result in also showing entities within a farther distance (e.g., 50 meters) at a lower frequency (e.g., 10 hertz). Similarly, for big things far away, a large number of components and reduced frequency updates may be used.
A sniper rifle in a game simulation may comprise an interest, such as:
This interest may also be on the player's entity. Normally the player might have the 54->query mapping described above. If, however, the player equips a sniper rifle, the player entity might be updated to have {54->Query, 2003->Cone}, and the system may send the client both results. A client could set the interest to be InCone(range, fov), such as for a client authoritative game with server cheat detection. A server cheat detector worker may subscribe to updates on both Interest and SniperRifle, and may verify that the range and fov are accurate for the equipped rifle. For a server authoritative game, the client may set the fields in PlayerControls (or have a client-authoritative SniperRifle component validated server side). A managed worker may update an Interest component for the player entity's SniperRifle component with the field of view and range. Interest may similarly be expressed for, for example, binoculars.
Interest for a mini-map may be expressed as, for example:
The mini-map interest may be added in addition to one or more of the components described above. Most of the game screen may be the view of the area around the player, with a focus on the view along the cone of the sniper rifle. In the corner of the screen, for example, the game client could draw a small radar-like version of the map that just shows where other players are in the area, such as in the form of a mini-map. Because this is a mini-map, the appearance of any players, for example, might not need to be streamed to the client. Rather, the player entity might only need the subset of data that is included in the MapRepresentation component. Interest for team information may be expressed in several ways. For example, a HasEqualComponent-like method may be added, such as:
HasEqualComponent may return true for a pair of entities if both entities have the named component, and the values are equal.
As another example, a managed worker may control the interest. The managed worker may obtain the value of the relevant gameplay component (e.g., the team component in this example) and may query for entities with that component equal to the team value. The managed worker may update the Interest component on the player entity with the IDs of the found entities:
In massively multiplayer online (MMO) systems, team or guild membership information may be used. For example, a managed worker may set GuildMemberInfo, which may have multiple fields. Additionally or alternatively, GuildMemberInfo may be split into two (or more) components. A first component may be a comparator, and a second component may be the information to share if the comparator satisfies a constraint.
For stealth systems, interest may be expressed for anything in, for example, a local environment which is not in stealth mode. The interest component may be authoritative on a managed worker to avoid hacked clients seeing stealthed players. For example, interest may be represented as:
ComponentEquals may return true if an entity has a component_id with the provided number, and if that component has the value provided. To handle teams or sides being able to see stealth, the developer may add another constraint for HasEqualComponent(2002), or the managed worker may add interest in particular entity_id's similar to the team information use case. Whether to add another constraint or the managed worker adding interest may depend on whether stealth is very common in the game or relatively uncommon.
Interest may consider bandwidth constraints. For example, a developer may perform a single component update to the interest of any entities where the player client is authoritative to set it to a lower frequency and/or smaller radius. This may be done at a fine grained granularity, because if the client detects that it is using too much bandwidth, it may reduce (e.g., gradually) the frequency or query set via a component update and/or via command to a managed worker (e.g., for a server-authoritative component). Alternatively, a player entity may have multiple components, which may correspond to the client types (e.g., a DesktopClientComponent, a MobileClientComponent, etc.). The interest queries for each of those may be tailored for their bandwidth limits. An advantage of this approach may be that a similar mechanism can be used for managed workers. For example, assume a non-player component usually has a view distance of 10 meters. 10 meters may be the default interest set. However, through tuning, a game developer may know that their non-player component AI may become too slow if it tries to consider more than 20 entities. If the worker finds itself aware of too many entities (e.g., 50 entities of another threshold number), the worker may dynamically reduce the view distance (e.g., to 9 meters, 5 meters, etc.).
Whilst the embodiments and aspects have been described in relation to virtual hardware servers, the methods and systems may also be used with other hardware or servers including local or physical servers.
Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are described as example implementations of the following claims.
This application claims priority to and is a continuation of U.S. application Ser. No. 16/806,608 filed on Mar. 2, 2022, entitled “Simulation Systems and Methods Using Query-Based Interest,” which claims priority to U.S. application Ser. No. 16/134,372, filed Sep. 18, 2018, entitled “Simulation Systems and Methods Using Query-Based Interest,” which claims priority to and is a continuation-in-part of U.S. application Ser. No. 16/054,038, filed Aug. 3, 2018, entitled “A Scalable Simulation System with Scalable Data Propagation,” which claims priority to and is a continuation-in-part of U.S. application Ser. No. 16/008,125, filed Jun. 14, 2018, entitled “Simulation Systems and Methods,” which claims priority to and is a continuation of U.S. application Ser. No. 15/361,874, filed Nov. 28, 2016, entitled “Simulation Systems and Methods,” which claims priority to U.S. provisional application No. 62/378,715, filed Aug. 24, 2016, entitled “Simulation Systems and Methods”, by Robert James Frederick Whitehead et al. Each of the prior applications is incorporated by reference herein in their entirety for all purposes.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5173939 | Abadi et al. | Dec 1992 | A |
| 6826523 | Guy et al. | Nov 2004 | B1 |
| 7076493 | Myers et al. | Jul 2006 | B2 |
| 7383253 | Tsimelzon et al. | Jun 2008 | B1 |
| 7853615 | Liu et al. | Dec 2010 | B2 |
| 8135018 | Kandekar et al. | Mar 2012 | B1 |
| 8392400 | Ransil et al. | Mar 2013 | B1 |
| 9058090 | Walsh et al. | Jun 2015 | B1 |
| 9239739 | Semin et al. | Jan 2016 | B2 |
| 10025892 | Whitehead et al. | Jul 2018 | B2 |
| 10146877 | Zolnieryk et al. | Dec 2018 | B1 |
| 10942930 | Kemme et al. | Mar 2021 | B2 |
| 20030177187 | Evine et al. | Sep 2003 | A1 |
| 20060053163 | Liu et al. | Mar 2006 | A1 |
| 20070198958 | Tang et al. | Aug 2007 | A1 |
| 20090216599 | Robusto et al. | Aug 2009 | A1 |
| 20100113159 | Chapman et al. | May 2010 | A1 |
| 20100175108 | Protas | Jul 2010 | A1 |
| 20100245086 | Nielsen et al. | Sep 2010 | A1 |
| 20120054172 | Agrawal et al. | Mar 2012 | A1 |
| 20120179855 | Tsirkin et al. | Jul 2012 | A1 |
| 20120197997 | Kulkarni et al. | Aug 2012 | A1 |
| 20120317624 | Monjas Llorente et al. | Dec 2012 | A1 |
| 20130304903 | Mick et al. | Nov 2013 | A1 |
| 20130325433 | Albano et al. | Dec 2013 | A1 |
| 20140157259 | Dow et al. | Jun 2014 | A1 |
| 20140157289 | Aarts et al. | Jun 2014 | A1 |
| 20140379775 | Korangy et al. | Dec 2014 | A1 |
| 20160293133 | Dutt | Oct 2016 | A1 |
| 20170180338 | Hanatani et al. | Jun 2017 | A1 |
| 20180048532 | Poort et al. | Feb 2018 | A1 |
| 20180060138 | Whitehead et al. | Mar 2018 | A1 |
| 20180060461 | Whitehead et al. | Mar 2018 | A1 |
| 20180060462 | Whitehead et al. | Mar 2018 | A1 |
| 20190045232 | Sharif-Ahmadi et al. | Feb 2019 | A1 |
| 20190377672 | Ante et al. | Dec 2019 | A1 |
| Number | Date | Country |
|---|---|---|
| 101005514 | Jul 2007 | CN |
| 101582850 | Nov 2009 | CN |
| 102103657 | Jun 2011 | CN |
| 103593516 | Feb 2014 | CN |
| 104182502 | Dec 2014 | CN |
| 104267676 | Jan 2015 | CN |
| 108351812 | Jul 2018 | CN |
| 9944130 | Sep 1999 | WO |
| 2009108726 | Sep 2009 | WO |
| 2014052464 | Apr 2014 | WO |
| 2017107569 | Jun 2017 | WO |
| Entry |
|---|
| Najaran, M.T., et al. “SPEX: Scalable Spatial Publish/Subscribe for Distributed Virtual Worlds without Borders” MMSys '14, Proceedings of 5th ACM Multimedia Systems Conf., pp. 127-138 (2014) available from <https://dl.acm.org/doi/abs/10.1145/2557642.2557655> (Year: 2014). |
| Gascon-Samson, J., et al. “Dynamoth: A Scalable Pub/Sub Middleware for Latency-Constrained Applications in the Cloud” IEEE 35th Int'l. Conf. on Distributed Computing Systems, pp. 486-496 (2015) (Year: 2015). |
| Shen, S., et al. “Area of Simulation: Mechanism and Architecture for Multi-Avatar Virtual Environments” ACM Trans. Multimedia Comput. Commun. Appl., vol. 12, No. 1 (2015) (Year: 2015). |
| Jul. 11, 2023 U.S. Notice of Allowance—U.S. Appl. No. 17/069,982 (pp. 1-35). |
| GameHorizon 2014—Games Development: A Different Path (Herman Narula, Improbable) http://www.youtube.com/watch?v=1fpaoKLxSE4 Jun. 3, 2014 (pp. 1-3). |
| Unity3D White Paper “A Guide to Moving From Internal Game Engine Technology” Oct. 16, 2015 (pp. 1-28). |
| Feb. 27, 2017—U.S. Office Action—U.S. Appl. No. 15/361,874 (pp. 1-11). |
| Apr. 21, 2017—U.S. Office Action—U.S. Appl. No. 15/361,874 (pp. 1-12). |
| Carlos Eduardo Benevides Bezerra et al., “A Load Balancing Scheme for Massively Multiplayer Online Games,” Multimedia Tools and Applications, Kluwer Academic Publishers, vol. 45, No. 1-3, May 21, 2009, pp. 263-270. |
| Masaki Kohana, “Dynamic Data Allocation Method for Web-Based Multiserver Systems,” Jan. 1, 2012. (pp. 1-74). |
| Roman Chertov et al., “Optimistic Load Balancing in a Distributed Virtual Environment,” Network and Operating Systems Support for Digital Audio and Video, May 26, 2006. (pp. 1-6). |
| Fengyun Lu et al., “Load Balancing for Massively Multiplayer Online Games, ” The 5th Workshop on Network & System Support for Games 2006—Netgames 2006. (pp. 1-11). |
| “What We Found When We Simulated the Backbone of the Entire Internet on SpatialOS” Published Mar. 24, 2016, https://improbable.io/company/news/2016/03/24/what-we-found-when-we-simulated-the-backbone-of-the-entire-internet-on-spatialos (pp. 1-10). |
| Jul. 4, 2017 (W0) International Search Report and Written Opinion—PCT/GB2017/051231 (pp. 1-11). |
| Feb. 12, 2018 U.S. Final Office Action—U.S. Appl. No. 15/361,874 (pp. 1-10). |
| May 17, 2018 U.S. Notice of Allowance—U.S. Appl. No. 15/361,874 (pp. 1-10). |
| Sep. 14, 2018—U.S. Non-Final Office Action—U.S. Appl. No. 15/361,921 (pp. 1-33). |
| Jan. 2, 2019—U.S. Non-Final Office Action—U.S. Appl. No. 15/361,889 (pp. 1-21). |
| Dec. 13, 2018—(CN) Office Action—App 201780038250 (pp. 1-20). |
| Carlos Eduardo Benevides Bezerra et al. “A load balancing scheme for massively multiplayer online games” Multimedia Tools and Applications, May 21, 2009 pp. 1-27. |
| Ta Nguyen Binh Duong et al. “A Dynamic Load Sharing Algorithm for Massively Multiplayer Online Games” School of Computer Engineering, Nanyang Technological University, Singapore 639798, IEEE 2003, pp. 131-136. |
| Dec. 5, 2018—U.S. Office Action—U.S. Appl. No. 15/361,921 (pp. 1-24). |
| Jan. 2, 2019—U.S. Office Action—U.S. Appl. No. 15/361,889 (pp. 1-21). |
| Feb. 21, 2019—U.S. Non-Final Office Action—U.S. Appl. No. 16/134,372 (pp. 1-25). |
| Morse, K., et al. “Interest Management in Large-Scale Virtual Environments” Presence, vol. 9, No. 1, pp. 52-68 (2000) (Year: 2000). |
| Ricci, L. & Carlini, E. “Distributed Virtual Environments: From Client Server to Cloud and P2P Architectures” IEEE Int'l Conf. on High Performance Computing & Simulation (2012) available from <https://ieeexplore.ieee.org/ document/6266885> (Year: 2012) (pp. 1-1). |
| Smed, J., T. Kaukoranta, & H. Hakonen, “A Review on Networking and Multiplayer Computer Games” TUCS Technical Report No. 454 (2002) (Year: 2002) (pp. 1-30). |
| Ali, A. et al. “An Overview of Networking Infrastructure for Massively Multiplayer Online Games” (2013) available at <https://pdfs.semanticscholar.org/887b/374b1010ca23a01bb89b71e40fd31e34ce92.pdf> (Year: 2013) (pp. 1-10). |
| Feb. 26, 2019—U.S. Non-Final Office Action—U.S. Appl. 16/054,102 (pp. 1-19). |
| Feb. 5, 2019 (PCT)—Search Report and Written Opinion—App. PCT/GB2018/053058 (pp. 1-13). |
| Mar. 18, 2019—U.S. Notice of Allowance—U.S. Appl. No. 15/361,921 (pp. 1-11). |
| Mar. 27, 2019—U.S. Notice of Allowance—U.S. Appl. No. 16/054,135 (pp. 1-22). |
| Feb. 28, 2019 (CN)—Second Office Action—App. 2017800028250 (pp. 1-8 of original document; pp. 1-10 of translated document). |
| May 16, 2019—U.S. Non-Final Office Action—U.S. Appl. No. 16/054,038 (pp. 1-27). |
| Jul. 9, 2019—U.S. Final Office Action—U.S. Appl. No. 16/054,102 (pp. 1-28). |
| May 20, 2019—(CN) Notice of Allowance—App. 201780003825.0 (pp. 1-1). |
| Aug. 21, 2019—U.S. Non-Final Office Action—U.S. Appl. No. 15/795,322 (pp. 1-32). |
| Aug. 21, 2019—U.S. Non-Final Office Action—U.S. Appl. No. 16/008,125—pp. 1-26. |
| Aug. 30, 2019—U.S. Notice of Allowance—U.S. Appl. No. 16/134,372 (pp. 1-22). |
| Sep. 26, 2019—U.S. Notice of Allowance—U.S. Appl. No. 16/054,038 (pp. 1-15). |
| Nov. 7, 2019—International Search Report and Written Opinion—App PCT/GB2019/052627, 19 pages. |
| Dec. 24, 2019—(CN) First Office Action—App 2019101057902. |
| Dec. 26, 2019—(CN) First Office Action—App 2019101052932. |
| Jul. 9, 2020—U.S. Non-Final Office Action—U.S. Appl. No. 16/054,102 (pp. 1-11). |
| Jul. 29, 2020 U.S. Notice of Allowance—U.S. Appl. No. 15/795,322 (pp. 1-8). |
| Jul. 22, 2020—Notice of Allowance—U.S. Appl. No. 16/008,125. |
| Nov. 27, 2020—U.S. Final Office Action—U.S. Appl. No. 16/054,102 (pp. 1-17). |
| Dec. 10, 2020—U.S. Non-Final Office Action—U.S. Appl. No. 16/204,483 (pp. 1-72). |
| Arthur Valadares et al. “Enabling Fine-Grained Load Balancing for Virtual Worlds with Distributed Simulation Engines,” 2014; 12 pages. |
| Nov. 23, 2020—(CN) Notice of Allowance—App No. 201910105790.2. |
| Jan. 7, 2021—U.S. Non-Final Office Action—U.S. Appl. No. 16/523,554 (pp. 1-43). |
| Oct. 6, 2021—U.S. Non-Final Office Action—U.S. Appl. No. 16/204,483 (pp. 1-14). |
| Boulanger, J., et al. “Comparing Interest Management Algorithms for Massively Multiplayer Games,” 5th Workshop on Network & System Support for Games, 2006. |
| Mar. 24, 2022—U.S. Non-Final Office Action—U.S. Appl. No. 16/806,608 (pp. 1-45). |
| Mar. 31, 2023—(CN) First Office Action—App No. 201910719294.6. |
| Number | Date | Country | |
|---|---|---|---|
| 20230119908 A1 | Apr 2023 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62378715 | Aug 2016 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16806608 | Mar 2020 | US |
| Child | 18082146 | US | |
| Parent | 16134372 | Sep 2018 | US |
| Child | 16806608 | US | |
| Parent | 15361874 | Nov 2016 | US |
| Child | 16008125 | US | |
| Parent | 15361874 | Nov 2016 | US |
| Child | PCT/GB2017/051231 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16054038 | Aug 2018 | US |
| Child | 16134372 | US | |
| Parent | 16008125 | Jun 2018 | US |
| Child | 16054038 | US | |
| Parent | PCT/GB2017/051231 | May 2017 | US |
| Child | 16054038 | US |
