Patent Grant
10942795
Computing devices can utilize communication networks to exchange data. Companies and organizations operate computer networks that interconnect a number of computing devices to support operations or to provide services to third parties. The computing systems can be located in a single geographic location or located in multiple, distinct geographic locations (e.g., interconnected via private or public communication networks). Specifically, data centers or data processing centers, herein generally referred to as a “data center,” may include a number of interconnected computing systems to provide computing resources to users of the data center. The data centers may be private data centers operated on behalf of an organization or public data centers operated on behalf, or for the benefit of, the general public.
To facilitate increased utilization of data center resources, virtualization technologies allow a single physical computing device to host one or more instances of virtual machines that appear and operate as independent computing devices to users of a data center. With virtualization, the single physical computing device can create, maintain, delete, or otherwise manage virtual machines in a dynamic manner. In turn, users can request computer resources from a data center, including single computing devices or a configuration of networked computing devices, and be provided with varying numbers of virtual machine resources.
In some scenarios, virtual machine instances may be configured according to a number of virtual machine instance types to provide specific functionality. For example, various computing devices may be associated with different combinations of operating systems or operating system configurations, virtualized hardware resources and software applications to enable a computing device to provide different desired functionalities, or to provide similar functionalities more efficiently. These virtual machine instance type configurations are often contained within a device image, which includes static data containing the software (e.g., the OS and applications together with their configuration and data files, etc.) that the virtual machine will run once started. The device image is typically stored on the disk used to create or initialize the instance. Thus, a computing device may process the device image in order to implement the desired software configuration.
Generally described, aspects of the present disclosure relate to managing task executions on an on-demand code execution system, which may also be referred to as a “serverless” execution system, in order to reduce execution times by increasing a probability that an execution is distributed to a host computing device with an execution environment pre-configured to handle that execution. Specifically, and as described in more detail below, embodiments of the present disclosure address the “cold start” problem in serverless computing by providing reserved, pre-warmed capacity in a serverless system that can be used to service calls to a specific set of code (a “task”). Moreover, the present disclosure provides a variety of routing mechanisms enabling a frontend of a serverless system to route calls within the system in a manner that increases the chances that the call is routed to a “warm” (e.g., pre-provisioned) environment, even in cases where reserved capacity for a task is exceeded.
As described herein, an on-demand code execution system enables rapid execution of source code, which may be supplied by users of the on-demand code execution system. For example, a user may submit a script in a specific programming language (e.g., the PYTHON™ language) that, when executed, implements network-based processing for a user-facing application (e.g., a mobile device “app”). The on-demand code execution system can then enable the user to submit “calls” to execute that script, at which point the system will securely execute the script to provide the desired functionality. Unlike some other network-based services, an on-demand code execution system can remove the need for a user to maintain or configure a computing device, either virtual or physical, to support code execution. Due to this lack of need for a user to maintain a device, on-demand code execution systems are sometimes referred to as “serverless” systems (though of course the on-demand code execution system itself, as opposed to individual users, may maintain servers to support code execution).
To facilitate rapid on-demand code execution, the system can maintain a variety of execution environments (e.g., virtual machine instances, software containers, etc.) pre-provisioned with software, such as an operating system, code libraries and the like, used during execution of code. When a request to execute code is received, the system can attempt to route the request to such a pre-provisioned environment, enabling the code to execute within the environment quickly, without delay caused by provisioning the environment with software necessary to execute code. However, due to the wide variety of code that the system might support, a pre-provisioned environment may not always be available. Moreover, given the distributed nature of the system, even if a pre-provisioned environment did exist, it may be a non-trivial task to locate that environment within a reasonable time (e.g., in less time than it would take to simply provision an environment anew). When a pre-provisioned environment cannot be located, a call to execute code can be delayed while an environment is provisioned with necessary software. This delay is sometimes referred to as a “cold start,” since no “warm” (e.g., pre-provisioned) environment was located for the call. Cold starts can add significant latency to calls to execute a task, and in some cases can inhibit adoption of the on-demand code execution system.
In some instances, the on-demand code execution system provisions environments based on past usage. For example, the first time a set of code representing a task is called, the call might result in a cold start. However, after execution of the code, the system may maintain the now-provisioned environment for a given period of time (e.g., 5 minutes). If another call to the task is received within that period, the system may attempt to route the call to the now-warm environment, avoiding cold starts. Thus, “steady state” traffic to the system (e.g., where a number of calls per second is relatively steady) may reduce the problem of cold starts. However, non-steady traffic may be more susceptible to cold starts. While the system may implement a variety of predictive techniques to attempt to pre-provision environments to handle non-steady traffic, in some cases users require a reduction in cold starts that cannot be guaranteed based on such predictive techniques.
To address this problem, embodiments of the present disclosure enable the use of reserved capacity on the on-demand code execution system, representing a set of execution environments that are maintained in a pre-provisioned state for code of a given task. For example, a user may require that the on-demand code execution system handle at least 500 concurrent executions of a task and the system may therefore maintain at least 500 execution environments pre-provisioned with software required by the code of the task, regardless of the actual number of calls to execute the task received at the system. It could therefore be expected that no cold starts would occur when the number of concurrent executions of the task does not exceed 500. However, to support many executions of many different tasks, the on-demand code execution system can generally operate in a distributed manner, such that many devices act cooperatively to implement each function of the system. For example, a “frontend” of the system, to which calls to execute a task are addressed, may in fact be implemented by a variety of different frontend devices. As is common in distributed systems, it may be impossible or impractical for each frontend to maintain perfect and up-to-date information on the state. Moreover, a number of different sets of reserved capacity may exist for different functions or different customers, and each set of reserved capacity may be distributed across a number of host devices. Still further, each frontend may be expected to handle not only calls for execution of tasks with reserved capacity, but also calls for execution of tasks without reserved capacity. It is therefore non-trivial to implement a routing mechanism that can be implemented largely independently at each frontend such that all calls to execute tasks—with or without reserved capacity—have a high probability of being routed to a pre-provisioned with software to execute code of the task.
Embodiments of the present disclosure provide a routing mechanism to achieve this goal. Specifically, as disclosed herein, execution environments of the on-demand code execution system can be managed by a distributed set of “worker managers,” representing devices that manage a set of execution environments hosted by host devices (“workers”). Each of a distributed set of frontends may implement a consistent hash ring routing mechanism, which divides calls to execute tasks among the worker managers, such that each call is routed to an individual worker manager selected from the ring based on a hash value associated with the call. For example, with respect to general calls (e.g., for tasks without reserved capacity), an account identifier of a calling entity may be hashed into a position within a hash ring, and the call can be routed to a worker manager associated with that position. Because the capacity of an individual worker manager is typically limited, it may be undesirable for each frontend to route all calls from a given account to the same worker manager. Instead, the frontend may select a number of worker managers among which to distribute calls associated with the account, based on a number of desired concurrent executions for the account supported by the on-demand code executions system (which number may generally be referred to herein as a “throttle limit” for the account). For example, the system may be configured such that each worker manager supports 500 concurrent executions of tasks associated with a given account. Thus, where an account is configured to support a throttle limit of 2000 concurrent task executions, execution calls associated with the account can be divided among four worker managers, which may be the worker manager associated with a position in the hash ring corresponding to a hash of an account identifier, plus the three subsequent worker managers on the ring. Because these four worker managers occupy a subset of the entire hash ring, the set of worker managers associated with a given group of task calls (e.g., calls associated with a given account) can generally be referred to as an “arc” of the group of calls. Each frontend may implement the same arc-creation logic, and thus the arc of a group of calls can be expected to be the same across all frontends.
In some embodiments, throttle limits and associated account arcs may be configured to scale based on load associated with the arc. For example, at each period of a set of periods (e.g., every 5 minutes), a frontend may increment or decrement a throttle limit and associated account arc based on an average number concurrent executions for the arc over a past period. Accordingly, when the load on an arc reaches a threshold percentage of total capacity (e.g., 80% of arc capacity, such as 500 concurrent execution capacity represent by an arc-length of one worker manager), the frontend may increase the throttle limit by 500, and also increase the arc length by adding a worker manager to the arc, thus increasing capacity of the arc by the number of concurrent executions provided by that worker manager. When the load on an arc drops below a threshold capacity (e.g., 60% of arc capacity, represented for example as 1000 execution capacity represented by an arc-length of two worker managers), the frontend may decrease the throttle limit by 500 executions, and also decrease and the arc length by removing a worker manager to the arc, thus decreasing capacity of the arc by the number of concurrent executions provided by that worker manager. Accordingly, the size of throttle limit and an account arc may scale to account for load on the arc.
When a call associated with arc is received at a frontend, the frontend can then distribute the call to a worker manager selected among those within the arc. While a variety of load-balancing techniques can be used within an arc, in one embodiment each frontend implements a round robin selection to select a worker manager within the arc. Grouping calls of each account into per-account arcs can beneficially reduce the resources required to implement each frontend, when compared to maintaining arcs for each task. Moreover, use of per-account arcs can provide greater efficiency of the system, as in some cases grouping of task executions for an account at specific worker managers can increase efficiency of operation of such worker managers. For example, the system may be configured such that different task executions associated with a given account may be implement on the same virtual machine instance (e.g., as different containers on the instance), whereas task executions of different accounts may be required to execute on different virtual machine instances (to heighten security). However, per-account arcs may also reduce the chance that a call for a given task is routed to a worker manager that manages a warm environment for the task. For example, where an arc of an account includes four worker managers and two iterative calls to execute a task are made to a given frontend, round robin selection at the frontend might result in a first worker manager managing the first call and a second worker manager managing the second call. As such, the second call may experience a cold start.
In the case of tasks with reserved capacity, to enable consistent routing of calls to environments reserved for those tasks, each frontend can be configured to maintain a distinct arc for each task with reserved capacity. Specifically, each frontend may treat calls to a task with reserved capacity as a group of calls, and thus assign to the group an arc within a consistent hash ring (e.g., a set of n worker managers selected based on a hash value of a function identifier as a position within the ring). The set of worker managers within the arc is illustratively selected based on a number of reserved environments of the task. For example, if a worker manager is configured to handle no more than 500 environments for a given arc, and a user has requested 800 reserved environments, the arc size can be set to two, with each worker manager handling an equal portion (400) of the environments. Each worker manager within the arc can be configured with access to their respective portion of reserved environments, such that when each frontend routes calls to execute the task to one of the two worker managers, the worker manager can allocate the call to a reserved environment for the task, thus enabling execution of the task without incurrence of cold starts. Unlike account arcs, reserved capacity arcs are in some configurations statically sized, such that the arc begins and continues at a specific length corresponding to the total reserved capacity.
Similarly, a task associated with reserved capacity may be associated with a throttle limit distinct from an account's throttle limit. The reserved capacity throttle limit can be set equal to the current reserved capacity (e.g., 500 environments) without respect to current use of that capacity.
Accordingly, when a call is received at a frontend, the frontend may determine the appropriate arc of the consistent hash ring to utilize in routing the call (e.g., an account arc or, where reserved capacity exists, a reserved capacity arc). The frontend may then select a worker manager associated with that arc, and distribute the call to the worker manager. The worker manager can then locate an environment for the call (preferably, a pre-provisioned environment) and execute the called task within that environment. Because calls to tasks with reserved capacity are routed to worker managers that maintain reserved environments for those calls, the probability that such calls are routed to a manager with a pre-warmed environment is very high (and in some cases, may be 100%). Thus, the routing mechanisms described herein enable each frontend to accurately route calls associated with reserved capacity, without interfering with routing of other calls.
Another problem that may arise from use of reserved concurrency in an on-demand code execution system is that of executions exceeding reserved capacity. One benefit of on-demand code execution is scalability—because a user need not configure and manage their own devices, the system enables the user to scale operations, simply calling for as many concurrent task executions as the user may require. However, when a user reserves a specific number of environments, it can be desirable that the system also handles calls beyond that number of environments. That is, simply because a user has reserved 500 environments, it may be undesirable for the system to reject a 501st call to execute a task. Moreover, it may generally be desirable that calls exceeding a throttle limit of reserved capacity are handled as would be other more general calls on the system, such as by being routed based on an account arc rather than a reserved capacity arc. Accordingly, each frontend in embodiments of the present disclosure can be configured to allow for handling of “spillover” calls—calls to a function with reserved capacity that would exceed that the throttle limit of the reserved capacity. Specifically, the on-demand code execution system may be provisioned with a “counting service” that enables each frontend to report executions associated with a specific throttle limit (e.g., a reserved capacity throttle limit or account throttle limit), and obtain a count of executions associated with that throttle limit across all frontends. When a frontend obtains a call to a task with reserved capacity that exceeds a throttle limit of that reserved capacity (e.g., a 501st call to a task with 500 reserved environments), the frontend can be configured to handle such a call as a general call from the account, rather than a call to reserved capacity.
One potential side effect of handling calls that exceed reserved capacity in this manner is the potential for cold-starts of such spillover calls. For example, where a 501st call to a task with 500 reserved environments is routed to a worker manager within a general arc, there is a substantial probability that this manager does not have an available pre-warmed environment for the call, and as such, the call may experience a cold start. While this may be acceptable in many situations (e.g., because the number of calls exceeds reserved capacity, and thus the user presumably accepts a possibility of cold starts), it is generally undesirable and in some instances can lead to unintuitive operation. For example, as discussed above, even without reserved capacity the on-demand code execution system can generally minimize cold starts under steady-state execution levels for a task, where the number of concurrent executions remains generally constant over time. Accordingly, if a user initially reserves capacity for a task but later recognizes that the task is experiencing steady-state execution levels, the user may attempt to reduce or revoke the reserved capacity (e.g., to reduce a cost of the user imposed by the system for reserved capacity, accounting for the computational costs required to maintain reserved capacity). However, where a frontend “spills over” calls that exceed reserved capacity to an account arc, reducing or revoking reserved capacity may cause all or a portion of those steady-state calls to spillover to the account arc and thus experience cold starts. This incurrence of cold starts due to reduction in reserved capacity under steady-state executions level represents undesirable operation. Moreover, these cold starts may in fact increase computational load of the system, as the cold starts require provisioning new environments on the account arc, despite the fact that pre-warmed environments exist on the reserved capacity arc.
To address this problem of cold starts due to reductions in reserved capacity, a frontend according to embodiments of the present disclosure may continue to route spillover calls to a reserved capacity arc so long as worker managers of that arc have warm (but not “hot,” e.g., currently in use) environments for the task. Specifically, each frontend may be configured to maintain, for an arc, load information for the arc indicating a proportion of environments of the arc that are in use at a worker manager, compared to warm but not in-use environments. When the frontend receives a request for a task that would otherwise be handled as a spillover request (e.g., exceeding a throttle limit for reserved capacity), the frontend can instead be configured to route the task based on a reserved capacity arc, so long as the load of the arc falls under a given threshold (e.g., 95%, 99%, etc.). Moreover, worker managers on the arc (and potentially generally) can be configured to release warm environments into a pool of available resources when those environments are not utilized for a threshold period (e.g., 5 minutes). Thus, even when reserved capacity and a corresponding throttle limit are reduced, requests for a task with now-reduced reserved capacity will continue to be routed to the reserved capacity arc up to the capacity of that arc (as established prior to reduction of the reserved capacity). Under steady-state execution levels, that capacity will remain constant and available to service request. However, if a number of concurrent executions (a “concurrency level”) for the task falls, worker managers of that arc may release pre-warmed environments thus causing a reduction in capacity of that arc. This configuration can enable “graceful” scale down of reserved capacity arcs, such that cold starts are not experience under steady-state concurrency levels.
The general operation of the on-demand code execution system will now be discussed. Specifically, the on-demand code execution system disclosed herein may enable users to create request-drive services, by submitting or designating computer-executable source code to be executed by virtual machine instances on the on-demand code execution system in response to user requests to access such services. Each set of code on the on-demand code execution system may define a “task,” and implement specific functionality corresponding to that task when executed on a virtual machine instance of the on-demand code execution system (e.g., functionality of a network-based service). Individual implementations of the task on the on-demand code execution system may be referred to as an “execution” of the task (or a “task execution”). The on-demand code execution system can further enable users to trigger execution of a task based on a variety of potential events, such as detecting new data at a network-based storage system, transmission of an application programming interface (“API”) call to the on-demand code execution system, or transmission of a specially formatted hypertext transport protocol (“HTTP”) packet to the on-demand code execution system. Thus, users may utilize the on-demand code execution system to execute any specified executable code “on-demand,” without requiring configuration or maintenance of the underlying hardware or infrastructure on which the code is executed. Further, the on-demand code execution system may be configured to execute tasks in a rapid manner (e.g., in under 100 milliseconds [ms]), thus enabling execution of tasks in “real-time” (e.g., with little or no perceptible delay to an end user). To enable this rapid execution, the on-demand code execution system can include one or more virtual machine instances that are “pre-warmed” or pre-initialized (e.g., booted into an operating system and executing a complete or substantially complete runtime environment) and configured to enable execution of user-defined code, such that the code may be rapidly executed in response to a request to execute the code, without delay caused by initializing the virtual machine instance. Thus, when an execution of a task is triggered, the code corresponding to that task can be executed within a pre-initialized virtual machine in a very short amount of time.
Specifically, to execute tasks, the on-demand code execution system described herein may maintain a pool of pre-initialized virtual machine instances that are ready for use as soon as a user request is received. Due to the pre-initialized nature of these virtual machines, delay (sometimes referred to as latency) associated with executing the user code (e.g., instance and language runtime startup time) can be significantly reduced, often to sub-100 millisecond levels. Illustratively, the on-demand code execution system may maintain a pool of virtual machine instances on one or more physical computing devices, where each virtual machine instance has one or more software components (e.g., operating systems, language runtimes, libraries, etc.) loaded thereon. When the on-demand code execution system receives a call to execute the program code of a user (a “task”), which specifies one or more computing constraints for executing the program code of the user, the on-demand code execution system may select a virtual machine instance for executing the program code of the user based on the one or more computing constraints specified by the request and cause the program code of the user to be executed on the selected virtual machine instance. The program codes can be executed in isolated containers that are created on the virtual machine instances. Since the virtual machine instances in the pool have already been booted and loaded with particular operating systems and language runtimes by the time the requests are received, the delay associated with finding compute capacity that can handle the requests (e.g., by executing the user code in one or more containers created on the virtual machine instances) is significantly reduced.
The on-demand code execution system may include a virtual machine instance manager (a “worker manager”) configured to receive user code (threads, programs, etc., composed in any of a variety of programming languages) and execute the code in a highly scalable, low latency manner, without requiring user configuration of a virtual machine instance. Specifically, the virtual machine instance manager can, prior to receiving the user code and prior to receiving any information from a user regarding any particular virtual machine instance configuration, create and configure virtual machine instances according to a predetermined set of configurations, each corresponding to any one or more of a variety of run-time environments. Thereafter, the virtual machine instance manager receives user-initiated requests to execute code, and identifies a pre-configured virtual machine instance to execute the code based on configuration information associated with the request. The virtual machine instance manager can further allocate the identified virtual machine instance to execute the user's code at least partly by creating and configuring containers inside the allocated virtual machine instance, and provisioning the containers with code of the task as well as dependency code objects. Various embodiments for implementing a virtual machine instance manager and executing user code on virtual machine instances is described in more detail in U.S. Pat. No. 9,323,556, entitled “PROGRAMMATIC EVENT DETECTION AND MESSAGE GENERATION FOR REQUESTS TO EXECUTE PROGRAM CODE” and filed Sep. 30, 2014 (“the '556 patent”), the entirety of which is hereby incorporated by reference.
As used herein, the term “virtual machine instance” is intended to refer to an execution of software or other executable code that emulates hardware to provide an environment or platform on which software may execute (an “execution environment”). Virtual machine instances are generally executed by hardware devices, which may differ from the physical hardware emulated by the virtual machine instance. For example, a virtual machine may emulate a first type of processor and memory while being executed on a second type of processor and memory. Thus, virtual machines can be utilized to execute software intended for a first execution environment (e.g., a first operating system) on a physical device that is executing a second execution environment (e.g., a second operating system). In some instances, hardware emulated by a virtual machine instance may be the same or similar to hardware of an underlying device. For example, a device with a first type of processor may implement a plurality of virtual machine instances, each emulating an instance of that first type of processor. Thus, virtual machine instances can be used to divide a device into a number of logical sub-devices (each referred to as a “virtual machine instance”). While virtual machine instances can generally provide a level of abstraction away from the hardware of an underlying physical device, this abstraction is not required. For example, assume a device implements a plurality of virtual machine instances, each of which emulate hardware identical to that provided by the device. Under such a scenario, each virtual machine instance may allow a software application to execute code on the underlying hardware without translation, while maintaining a logical separation between software applications running on other virtual machine instances. This process, which is generally referred to as “native execution,” may be utilized to increase the speed or performance of virtual machine instances. Other techniques that allow direct utilization of underlying hardware, such as hardware pass-through techniques, may be used, as well.
While a virtual machine executing an operating system is described herein as one example of an execution environment, other execution environments are also possible. For example, tasks or other processes may be executed within a software “container,” which provides a runtime environment without itself providing virtualization of hardware. Containers may be implemented within virtual machines to provide additional security, or may be run outside of a virtual machine instance.
The foregoing aspects and many of the attendant advantages of this disclosure will become more readily appreciated as the same become better understood by reference to the following description, when taken in conjunction with the accompanying drawings.
The illustrative environment 100 further includes one or more auxiliary services 106, which can interact with the one-demand code execution environment 110 to implement desired functionality on behalf of a user. Auxiliary services 106 can correspond to network-connected computing devices, such as servers, which generate data accessible to the one-demand code execution environment 110 or otherwise communicate to the on-demand code execution environment 110. For example, the auxiliary services 106 can include web services (e.g., associated with the user computing devices 102, with the on-demand code execution system 110, or with third parties), databases, really simple syndication (“RSS”) readers, social networking sites, or any other source of network-accessible service or data source. In some instances, auxiliary services 106 may be invoked by code execution on the on-demand code execution system 110, such as by API calls to the auxiliary services 106. In some instances, auxiliary services 106 may be associated with the on-demand code execution system 110, e.g., to provide billing or logging services to the on-demand code execution system 110. In some instances, auxiliary services 106 actively transmit information, such as API calls or other task-triggering information, to the on-demand code execution system 110. In other instances, auxiliary services 106 may be passive, such that data is made available for access by the on-demand code execution system 110. For example, components of the on-demand code execution system 110 may periodically poll such passive data sources, and trigger execution of tasks within the on-demand code execution system 110 based on the data provided. While depicted in
The illustrative environment 100 further includes one or more network-based data storage services 108, configured to enable the on-demand code execution system 110 to store and retrieve data from one or more persistent or substantially persistent data sources. Illustratively, the network-based data storage services 108 may enable the on-demand code execution system 110 to store information corresponding to a task, such as source code or metadata, to store additional code objects representing dependencies of tasks, to retrieve data to be processed during execution of a task, and to store information (e.g., results) regarding that execution. The network-based data storage services 108 may represent, for example, a relational or non-relational database. In another example, the network-based data storage services 108 may represent a network-attached storage (NAS), configured to provide access to data arranged as a file system. The network-based data storage services 108 may further enable the on-demand code execution system 110 to query for and retrieve information regarding data stored within the on-demand code execution system 110, such as by querying for a number of relevant files or records, sizes of those files or records, file or record names, file or record creation times, etc. In some instances, the network-based data storage services 108 may provide additional functionality, such as the ability to separate data into logical groups (e.g., groups associated with individual accounts, etc.). While shown as distinct from the auxiliary services 106, the network-based data storage services 108 may in some instances also represent a type of auxiliary service 106.
The user computing devices 102, auxiliary services 106, and network-based data storage services 108 may communicate with the on-demand code execution system 110 via network 104, which may include any wired network, wireless network, or combination thereof. For example, the network 104 may be a personal area network, local area network, wide area network, over-the-air broadcast network (e.g., for radio or television), cable network, satellite network, cellular telephone network, or combination thereof. As a further example, the network 104 may be a publicly accessible network of linked networks, possibly operated by various distinct parties, such as the Internet. In some embodiments, the network 104 may be a private or semi-private network, such as a corporate or university intranet. The network 104 may include one or more wireless networks, such as a Global System for Mobile Communications (GSM) network, a Code Division Multiple Access (CDMA) network, a Long Term Evolution (LTE) network, or any other type of wireless network. The network 104 can use protocols and components for communicating via the Internet or any of the other aforementioned types of networks. For example, the protocols used by the network 104 may include Hypertext Transfer Protocol (HTTP), HTTP Secure (HTTPS), Message Queue Telemetry Transport (MQTT), Constrained Application Protocol (CoAP), and the like. Protocols and components for communicating via the Internet or any of the other aforementioned types of communication networks are well known to those skilled in the art and, thus, are not described in more detail herein.
The on-demand code execution system 110 is depicted in
Further, the on-demand code execution system 110 may be implemented directly in hardware or software executed by hardware devices and may, for instance, include one or more physical or virtual servers implemented on physical computer hardware configured to execute computer executable instructions for performing various features that will be described herein. The one or more servers may be geographically dispersed or geographically co-located, for instance, in one or more data centers. In some instances, the one or more servers may operate as part of a system of rapidly provisioned and released computing resources, often referred to as a “cloud computing environment.”
In the example of
In
To enable interaction with the on-demand code execution system 110, the system 110 includes multiple frontends 120, which enable interaction with the on-demand code execution system 110. In an illustrative embodiment, the frontends 120 serve as a “front door” to the other services provided by the on-demand code execution system 110, enabling users (via user computing devices 102) to provide, request execution of, and view results of computer executable source code. The frontends 120 include a variety of components to enable interaction between the on-demand code execution system 110 and other computing devices. For example, each frontend 120 may include a request interface providing user computing devices 102 with the ability to upload or otherwise communication user-specified code to the on-demand code execution system 110 and to thereafter request execution of that code. In one embodiment, the request interface communicates with external computing devices (e.g., user computing devices 102, auxiliary services 106, etc.) via a graphical user interface (GUI), CLI, or API. The frontends 120 process the requests and makes sure that the requests are properly authorized. For example, the frontends 120 may determine whether the user associated with the request is authorized to access the source code specified in the request.
References to source code as used herein may refer to any program code (e.g., a program, routine, subroutine, thread, etc.) written in a specific program language. In the present disclosure, the terms “source code,” “user code,” and “program code,” may be used interchangeably. Source code which has been compiled for execution on a specific device is generally referred to herein as “machine code.” Both “source code” and “machine code” are representations of the same instructions, which may be collectively referred to as “code.” Such code may be executed to achieve a specific function, for example, in connection with a particular web application or mobile application developed by the user. As noted above, individual collections of code (e.g., to achieve a specific function) are referred to herein as “tasks,” while specific executions of that code are referred to as “task executions” or simply “executions.” Source code for a task may be written, by way of non-limiting example, in JavaScript (e.g., node.js), Java, Python, and/or Ruby (and/or another programming language). Tasks may be “triggered” for execution on the on-demand code execution system 110 in a variety of manners. In one embodiment, a user or other computing device may transmit a request to execute a task may, which can generally be referred to as “call” to execute of the task. Such calls may include the source code (or the location thereof) to be executed and one or more arguments to be used for executing the source code. For example, a call may provide the source code of a task along with the request to execute the task. In another example, a call may identify a previously uploaded task by its name or an identifier. In yet another example, source code corresponding to a task may be included in a call for the task, as well as being uploaded in a separate location (e.g., storage of an auxiliary service 106 or a storage system internal to the on-demand code execution system 110) prior to the request being received by the on-demand code execution system 110. A request interface of the frontend 120 may receive calls to execute tasks as Hypertext Transfer Protocol Secure (HTTPS) requests from a user. Also, any information (e.g., headers and parameters) included in the HTTPS request may also be processed and utilized when executing a task. As discussed above, any other protocols, including, for example, HTTP, MQTT, and CoAP, may be used to transfer the message containing a task call to the request interface 122.
A call to execute a task may specify one or more third-party libraries (including native libraries) to be used along with the user code corresponding to the task. In one embodiment, the call may provide to the on-demand code execution system 110 a ZIP file containing the source code and any libraries (and/or identifications of storage locations thereof) corresponding to the task requested for execution. In some embodiments, the call includes metadata that indicates the source code of the task to be executed, the language in which the source code is written, the user associated with the call, and/or the computing resources (e.g., memory, etc.) to be reserved for executing the source code. For example, the source code of a task may be provided with the call, previously uploaded by the user, provided by the on-demand code execution system 110 (e.g., standard routines), and/or provided by third parties. Illustratively, code not included within a call or previously uploaded by the user may be referenced within metadata of the task by use of a URI associated with the code. In some embodiments, such resource-level constraints (e.g., how much memory is to be allocated for executing a particular code) are specified for the particular task, and may not vary over each execution of the task. In such cases, the on-demand code execution system 110 may have access to such resource-level constraints before each individual call is received, and the individual call may not specify such resource-level constraints. In some embodiments, the call may specify other constraints such as permission data that indicates what kind of permissions or authorities that the call invokes to execute the task. Such permission data may be used by the on-demand code execution system 110 to access private resources (e.g., on a private network). In some embodiments, individual code sets may also be associated with permissions or authorizations. For example, a third party may submit a code object and designate the object as readable by only a subset of users. The on-demand code execution system 110 may include functionality to enforce these permissions or authorizations with respect to code sets.
In some embodiments, a call may specify the behavior that should be adopted for handling the call. In such embodiments, the call may include an indicator for enabling one or more execution modes in which to execute the task referenced in the call. For example, the call may include a flag or a header for indicating whether the task should be executed in a debug mode in which the debugging and/or logging output that may be generated in connection with the execution of the task is provided back to the user (e.g., via a console user interface). In such an example, the on-demand code execution system 110 may inspect the call and look for the flag or the header, and if it is present, the on-demand code execution system 110 may modify the behavior (e.g., logging facilities) of the container in which the task is executed, and cause the output data to be provided back to the user. In some embodiments, the behavior/mode indicators are added to the call by the user interface provided to the user by the on-demand code execution system 110. Other features such as source code profiling, remote debugging, etc. may also be enabled or disabled based on the indication provided in a call.
To manage requests for code execution, the frontend 120 can include an execution queue (not shown in
As noted above, tasks may be triggered for execution at the on-demand code execution system 110 based on explicit calls from user computing devices 102 (e.g., as received at the request interface 122). Alternatively or additionally, tasks may be triggered for execution at the on-demand code execution system 110 based on data retrieved from one or more auxiliary services 106 or network-based data storage services 108. To facilitate interaction with auxiliary services 106, the frontend 120 can include a polling interface (not shown in
In addition to tasks executed based on explicit user calls and data from auxiliary services 106, the on-demand code execution system 110 may in some instances operate to trigger execution of tasks independently. For example, the on-demand code execution system 110 may operate (based on instructions from a user) to trigger execution of a task at each of a number of specified time intervals (e.g., every 10 minutes).
The frontend 120 can further includes an output interface (not shown in
In accordance with embodiments of the present disclosure, each frontend illustratively includes a call distributor 122, which acts to identify a specific worker manager 140 to which to route a call to execute a task. Routines and interactions associated with the call distributor are described in more detail below; however, in brief, a call distributor 122 may distribute calls among worker managers 140 utilizing a consistent hash ring algorithm, in which calls are grouped into related sets (e.g., calls associated with an account, with reserved concurrency, etc.), and each set is associated with a group of worker managers 140 (an “arc”) selected from a circular organization of such managers (the “hash ring”). For each call, the distributor 122 may identify an arc associated with the call, and then distribute the call to a worker manager of the arc (e.g., in a round robin fashion among such worker managers). In instances where spillover of a reserved capacity arc occurs, the distributor 122 may route the call based either on the reserved capacity arc or on the account arc, based on remaining capacity of the account arc, in order to avoid cold starts that may be otherwise associated with spillover.
To facilitate operation of the call distributor 122, the system 110 further includes a counting service 170, representing a distributed network-accessible service configured to maintain a count of current executions associated with each throttle limit on the on-demand code execution system 110 (e.g. a throttle limit of an account, of reserved capacity, etc.). Specifically, the counting service 170 may provide interfaces such as an API including “countUp” and “countDown” functions, callable by each distributor 122 to increment or decrement a count associated with a throttle limit (e.g., at receiving a call and completing execution of the called task, respectively). The service 170, in response to a countUp or countDown call, or in response to another function (such as a “getCount” function) may return to a calling distributor 122 a count of the throttle limit, aggregated across all distributors 122 of all frontends 120. To facilitate scaling, the counting service 170 may operate in a distributed manner, such as by a number of instances of the service 170 each corresponding to a computing device (physical or virtual on physical) configured to implement an instance of the service 170. The instances may communicate with one another in an attempt to unify counts across the instances, and may utilize a consensus protocol, such as a Paxos-family protocol, to provide a current count. Given the distributed nature of the service 170 and the potential frequency of calls to a given arc, it is contemplated that the count provided by the service 170 for a throttle limit may not always be perfectly accurate. However, it is notably that embodiments of the present disclose enable the call distributor 122 to operate without requiring perfect knowledge of a concurrency level for a given throttle limit.
To execute tasks, the on-demand code execution system 110 includes one or more warming pool managers 130, which “pre-warm” (e.g., initialize) virtual machine instances in an attempt to enable tasks to be executed quickly, without the delay caused by initialization of the virtual machines. The on-demand code execution system 110 further includes one or more worker managers 140, which manage active virtual machine instances (e.g., currently assigned to execute tasks in response to task calls). While not shown in
The warming pool managers 130 can generally be configured to attempt to ensure that virtual machine instances are ready to be used by the worker managers 140 when the on-demand code execution system 110 detects an event triggering execution of a task on the on-demand code execution system 110. In the example illustrated in
As shown in
In some embodiments, the virtual machine instances in a warming pool 130A may be used to serve any user's calls. In one embodiment, all the virtual machine instances in a warming pool 130A are configured in the same or substantially similar manner. In another embodiment, the virtual machine instances in a warming pool 130A may be configured differently to suit the needs of different users. For example, the virtual machine instances may have different operating systems, different language runtimes, and/or different libraries loaded thereon. In yet another embodiment, the virtual machine instances in a warming pool 130A may be configured in the same or substantially similar manner (e.g., with the same OS, language runtimes, and/or libraries), but some of those instances may have different container configurations. For example, one instance might have a container created therein for running code written in Python, and another instance might have a container created therein for running code written in Ruby.
The warming pool managers 130 may pre-configure the virtual machine instances in a warming pool 130A, such that each virtual machine instance is configured to satisfy at least one of the operating conditions that may be requested or specified by a user when defining a task. In one embodiment, the operating conditions may include program languages in which the potential source code of a task may be written. For example, such languages may include Java, JavaScript, Python, Ruby, and the like. In some embodiments, the set of languages that the source code of a task may be written in may be limited to a predetermined set (e.g., set of 4 languages, although in some embodiments sets of more or less than four languages are provided) in order to facilitate pre-initialization of the virtual machine instances that can satisfy calls to execute the task. For example, when the user is configuring a task via a user interface provided by the on-demand code execution system 110, the user interface may prompt the user to specify one of the predetermined operating conditions for executing the task. In another example, the service-level agreement (SLA) for utilizing the services provided by the on-demand code execution system 110 may specify a set of conditions (e.g., programming languages, computing resources, etc.) that tasks should satisfy, and the on-demand code execution system 110 may assume that the tasks satisfy the set of conditions in handling the requests. In another example, operating conditions specified by a task may include: the amount of compute power to be used for executing the task; the type of triggering event for a task (e.g., an API call, HTTP packet transmission, detection of a specific data at an auxiliary service 106); the timeout for the task (e.g., threshold time after which an execution of the task may be terminated); and security policies (e.g., may control which instances in the warming pools 130A are usable by which user), among other specified conditions.
One or more worker managers 140 manage the instances used for servicing incoming calls to execute tasks. In the example illustrated in
As shown in
In the example illustrated in
Once a triggering event to execute a task has been successfully processed by a frontend 120, the frontend 120 passes a request to a worker manager 140 to execute the task. On receiving a request to execute a task, a worker manager 140 finds capacity to execute a task on the on-demand code execution system 110. For example, if there exists a particular virtual machine instance in the active pool 140A that has a container with the user code of the task already loaded therein (e.g., code 156D-1 shown in the container 156D) (e.g., a “warm” container), the worker manager 140 may assign the container to the task and cause the task to be executed in the container. Alternatively, if the user code of the task is available in the local cache of one of the virtual machine instances (e.g., codes 158G, 158H, which are stored on the instance 158 but do not belong to any individual containers), the worker manager 140 may create a new container on such an instance, assign the container to the task, and cause the user code of the task to be loaded and executed in the container.
If the worker manager 140 determines that the source code associated with the triggered task is not found on any of the instances (e.g., either in a container or the local cache of an instance) in the active pool 140A, the worker manager 140 may determine whether any of the instances in the active pool 140A is currently assigned to the user associated with the triggered task and has compute capacity to handle the triggered task. If there is such an instance, the worker manager 140 may create a new container on the instance and assign the container to execute the triggered task. Alternatively, the worker manager 140 may further configure an existing container on the instance assigned to the user, and assign the container to the triggered task. For example, the worker manager 140 may determine that the existing container may be used to execute the task if a particular library demanded by the task is loaded thereon. In such a case, the worker manager 140 may load the particular library and the code of the task onto the container and use the container to execute the task.
If the active pool 140 does not contain any instances currently assigned to the user, the worker manager 140 pulls a new virtual machine instance from the warming pool 130A, assigns the instance to the user associated with the triggered task, creates a new container on the instance, assigns the container to the triggered task, and causes the source code of the task to be downloaded and executed on the container.
In some embodiments, the on-demand code execution system 110 is adapted to begin execution of a task shortly after it is received (e.g., by the frontend 120). A time period can be determined as the difference in time between initiating execution of the task (e.g., in a container on a virtual machine instance associated with the user) and detecting an event that triggers execution of the task (e.g., a call received by the frontend 120). The on-demand code execution system 110 is adapted to begin execution of a task within a time period that is less than a predetermined duration. In one embodiment, the predetermined duration is 500 ms. In another embodiment, the predetermined duration is 300 ms. In another embodiment, the predetermined duration is 100 ms. In another embodiment, the predetermined duration is 50 ms. In another embodiment, the predetermined duration is 10 ms. In another embodiment, the predetermined duration may be any value chosen from the range of 10 ms to 500 ms. In some embodiments, the on-demand code execution system 110 is adapted to begin execution of a task within a time period that is less than a predetermined duration if one or more conditions are satisfied. For example, the one or more conditions may include any one of: (1) the source code of the task is loaded on a container in the active pool 140 at the time the request is received; (2) the source code of the task is stored in the code cache of an instance in the active pool 140 at the time the call to the task is received; (3) the active pool 140A contains an instance assigned to the user associated with the call at the time the call is received; or (4) the warming pool 130A has capacity to handle the task at the time the event triggering execution of the task is detected.
Once the worker manager 140 locates one of the virtual machine instances in the warming pool 130A that can be used to execute a task, the warming pool manager 130 or the worker manger 140 takes the instance out of the warming pool 130A and assigns it to the user associated with the request. The assigned virtual machine instance is taken out of the warming pool 130A and placed in the active pool 140A. In some embodiments, once the virtual machine instance has been assigned to a particular user, the same virtual machine instance cannot be used to execute tasks of any other user. This provides security benefits to users by preventing possible co-mingling of user resources. Alternatively, in some embodiments, multiple containers belonging to different users (or assigned to requests associated with different users) may co-exist on a single virtual machine instance. Such an approach may improve utilization of the available compute capacity.
In some embodiments, the on-demand code execution system 110 may maintain a separate cache in which code of tasks are stored to serve as an intermediate level of caching system between the local cache of the virtual machine instances and the account data store 164 (or other network-based storage not shown in
After the task has been executed, the worker manager 140 may tear down the container used to execute the task to free up the resources it occupied to be used for other containers in the instance. Alternatively, the worker manager 140 may keep the container running to use it to service additional calls from the same user. For example, if another call associated with the same task that has already been loaded in the container, the call can be assigned to the same container, thereby eliminating the delay associated with creating a new container and loading the code of the task in the container. In some embodiments, the worker manager 140 may tear down the instance in which the container used to execute the task was created. Alternatively, the worker manager 140 may keep the instance running to use it to service additional calls from the same user. The determination of whether to keep the container and/or the instance running after the task is done executing may be based on a threshold time, the type of the user, average task execution volume of the user, and/or other operating conditions. For example, after a threshold time has passed (e.g., 5 minutes, 30 minutes, 1 hour, 24 hours, 30 days, etc.) without any activity (e.g., task execution), the container and/or the virtual machine instance is shutdown (e.g., deleted, terminated, etc.), and resources allocated thereto are released. In some embodiments, the threshold time passed before a container is torn down is shorter than the threshold time passed before an instance is torn down.
In accordance with embodiments of the present disclosure, the warming pool managers 130 and worker managers 140 may collectively operate to ensure that, for a given task associated with reserved capacity, a number of execution environments (e.g., VM instances or containers on VM instances) provisioned for the task across managers 130 and 140 does not fall below the level of reserved capacity. These environments may thus be excluded from typical “tear down” policies of the managers 130 and 140, such that they are destroyed even when unused. Illustratively, the worker managers 130 may, on receiving a request for reserved capacity, create a required number of environments within the warming pool 130A, and may “vend” such environments to different worker managers 140 based on requests for the environments. When environments go unused at a given worker manager 140, the managers 140 “lease” on the environment may be expired, and the environment will revert to management by the set of warming pool managers 130. Thus, the managers 130 and 140 can ensure that an environment is available to service the first n requests for a task, where n is the reserved capacity of the task. In some instances, the number of environments maintained for a set of reserved capacity may exceed the reserved capacity by a minor amount (e.g., 1%, 5%, 10%, etc.). This may enable the system 110 to avoid cold starts that may otherwise occur (e.g., due to a difficulty of maintaining perfect state information in a distributed system) when a number of calls obtained is nearing the reserved capacity.
In some embodiments, the on-demand code execution system 110 may provide data to one or more of the auxiliary services 106 as it executes tasks in response to triggering events. For example, the frontends 120 may communicate with the monitoring/logging/billing services included within the auxiliary services 106. The monitoring/logging/billing services may include: a monitoring service for managing monitoring information received from the on-demand code execution system 110, such as statuses of containers and instances on the on-demand code execution system 110; a logging service for managing logging information received from the on-demand code execution system 110, such as activities performed by containers and instances on the on-demand code execution system 110; and a billing service for generating billing information associated with executing user code on the on-demand code execution system 110 (e.g., based on the monitoring information and/or the logging information managed by the monitoring service and the logging service). In addition to the system-level activities that may be performed by the monitoring/logging/billing services (e.g., on behalf of the on-demand code execution system 110), the monitoring/logging/billing services may provide application-level services on behalf of the tasks executed on the on-demand code execution system 110. For example, the monitoring/logging/billing services may monitor and/or log various inputs, outputs, or other data and parameters on behalf of the tasks being executed on the on-demand code execution system 110.
In some embodiments, the worker managers 140 may perform health checks on the instances and containers managed by the worker managers 140 (e.g., those in a corresponding active pool 140A). For example, the health checks performed by a worker manager 140 may include determining whether the instances and the containers managed by the worker manager 140 have any issues of (1) misconfigured networking and/or startup configuration, (2) exhausted memory, (3) corrupted file system, (4) incompatible kernel, and/or any other problems that may impair the performance of the instances and the containers. In one embodiment, a worker manager 140 performs the health checks periodically (e.g., every 5 minutes, every 30 minutes, every hour, every 24 hours, etc.). In some embodiments, the frequency of the health checks may be adjusted automatically based on the result of the health checks. In other embodiments, the frequency of the health checks may be adjusted based on user requests. In some embodiments, a worker manager 140 may perform similar health checks on the instances and/or containers in a warming pool 130A. The instances and/or the containers in a warming pool 130A may be managed either together with those instances and containers in an active pool 140A or separately. In some embodiments, in the case where the health of the instances and/or the containers in a warming pool 130A is managed separately from an active pool 140A, a warming pool manager 130, instead of a worker manager 140, may perform the health checks described above on the instances and/or the containers in a warming pool 130A.
In the depicted example, virtual machine instances (“instances”) 152, 154 are shown in a warming pool 130A managed by a warming pool manager 130, and instances 156, 158 are shown in an active pool 140A managed by a worker manager 140. The illustration of the various components within the on-demand code execution system 110 is logical in nature and one or more of the components can be implemented by a single computing device or multiple computing devices. For example, the instances 152, 154, 156, 158 can be implemented on one or more physical computing devices in different various geographic regions. Similarly, each frontend 120, warming pool manager 130, and worker manager 140 can be implemented across multiple physical computing devices. Alternatively, one or more of a frontend 120, a warming pool manager 130, and a worker manager 140 can be implemented on a single physical computing device. Although four virtual machine instances are shown in the example of
While some functionalities are generally described herein with reference to an individual component of the on-demand code execution system 110, other components or a combination of components may additionally or alternatively implement such functionalities. For example, frontends 120 may operate in a peer-to-peer manner to provide functionality corresponding to the counting service 170, as opposed to including a distinct service 170.
The memory 280 may contain computer program instructions (grouped as modules in some embodiments) that the processing unit 290 executes in order to implement one or more aspects of the present disclosure. The memory 280 generally includes random access memory (RAM), read only memory (ROM) and/or other persistent, auxiliary or non-transitory computer readable media. The memory 280 may store an operating system 284 that provides computer program instructions for use by the processing unit 290 in the general administration and operation of the code analysis system 160. The memory 280 may further include computer program instructions and other information for implementing aspects of the present disclosure. For example, in one embodiment, the memory 280 includes a user interface unit 282 that generates user interfaces (and/or instructions therefor) for display upon a computing device, e.g., via a navigation and/or browsing interface such as a browser or application installed on the computing device. In addition, the memory 280 may include and/or communicate with one or more data repositories (not shown), for example, to access user program codes and/or libraries.
In addition to and/or in combination with the user interface unit 282, the memory 280 may include a call distributor 122 that may be executed by the processing unit 290. In one embodiment, the call distributor 122 implements various aspects of the present disclosure (e.g., the routines of
In some embodiments, the frontend 120 may further include components other than those illustrated in
With reference to
The interactions of
In some instances, a user may request reserved capacity for a task, but request that only some calls to the task utilize such reserved capacity. Accordingly, in one embodiment, the system 110 may enable creation of multiple “copies” of a task, each associated with a different identifier (or “alias”). Accordingly, calls to a first alias of a task may utilize reserved capacity, while calls to a second alias may not.
At (2), frontend 120 transmits a call to the counting service 170 to increment a count associated with a throttle limit for the currently called task. For example, where the call is to a task with reserved capacity, the frontend 120 can increment a count associated with a throttle limit for the reserved capacity. Where the call is not associated with reserved capacity, the frontend 120 can increment a count associated with a throttle limit for an account under which the call will be executed. The service 170, in response, returns a current count associated with that throttle limit to the frontend 120, at (3). For ease of description, it will be assumed that this count does not exceed the associated throttle limit. However, interactions for handling calls that exceed an associated throttle limit will be discussed in more detail below.
Thereafter, at (4), frontend 120 identifies an arc of a consistent hash ring to utilize in routing the call. An example of such a ring is shown in
On obtaining a position within the ring 400, the frontend 120 can select an initial worker manager 140 to be associated with an arc of the call (the arc representing all worker managers 140 to which calls within a group of calls can be distributed). In one embodiment, the frontend 120 selects the worker manager 140 associated with first position of a manager 140 in the ring subsequent to the call's position on the ring (as set by the call's hash value). In other embodiment, the frontend 120 selects another the worker manager 140 in the ring based on the call's position on the ring (e.g., a first manager 140 prior to the call's position), though notably this could also be represented as a shift in position of managers 140 in the ring. For example, where a group of call's hash value (e.g., hash of an account identifier) falls between points 402 and 404, the group could be assigned an arc beginning with “WM2,” an illustrative manager 140 associated with point 404. Similarly, where a group of call's hash value (e.g., hash of an account identifier) falls between points 404 and 406, the group could be assigned an arc beginning with “WM3,” an illustrative manager 140 associated with point 406, etc. Because the hash values of each call group can be expected to be relatively random and distributed among the range 400, the initial worker manager 140 of each arc can also be expected to be evenly distributed across the managers 140. Moreover, the ring 400 enables consistent routing, such that calls within a given group consistently result in the same hash value. As a result, the likelihood that a call is routed to a manager 140 with an available warm environment increases relative to inconsistent routing, such as randomized routing.
As discussed above, in some instances each worker manager 140 can be configured with a threshold capacity for a given group of calls. For example, each manager 140 may be configured to have a capacity of 500 for each group of calls. Accordingly, where the supported concurrency (e.g., throttle limit) of group of calls exceeds the capacity of an individual manager 140, multiple managers may be assigned to the group of calls. Illustratively, the number of managers may be calculated as the supported concurrency divided by the concurrency capacity of an individual manager 140. In some instances, the supported concurrency may scale. For example, the system 110 may be configured to initially support up to 500 calls for a given group (e.g., an account identifier), and may scale up that supported concurrency over time (e.g., by increasing the supported concurrency by 500 calls for each 5 minute period that used concurrency exceeds a threshold percentage of supported concurrency). In other instances, supported concurrency may be static. For example, the supported concurrency for a group of calls corresponding to a task with reserved capacity may be set at the level of reserved capacity.
The set of worker managers 140 supporting calls within a given group illustratively represents an “arc” of that group, which may correspond to the initial manager 140 selected for the group based on a hash value of the group (as discussed above), as well as any additional worker managers 140 required to provide the supported concurrency. For example, if a given group had a supported concurrency of 1000 and each manager 140 had a capacity of 500 concurrent executions for the group, the arc length of the group would be 2. If the hash value of the group resulted in initial selection of WM5 in the ring 400 (e.g., because the value is located between points 408 and 410 on the ring), the arc of the group would consist of managers 140 WM5 and WM6. For groups without reserved capacity, it may be beneficial to maintain an arc at a shortest length possible to support the supported concurrency of the arc, as providing a larger arc may further distribute calls within the group and increase chances that a call lands on a manager 140 without an available warm environment. For groups with reserved capacity, the arc length may be set to the maximum of the reserved capacity, as each manager 140 within that arc can be expected to have available a reserved warm environment for the calls.
An illustrative table 420 of arcs is shown in
Returning to
Thereafter, a (5), the frontend 120 selects a worker manager 140 associated with the arc to which to distribute the call. Illustratively, the frontend 120 may select the manager 140 from among those in the arc in a “round robin” fashion, selecting a next manager 140 in the arc relative to a manager 140 in the arc selected for a past call of the arc.
On selecting a manager 140, the frontend 120 then distributes the call to the manager 140, at (6). The manager 140, in turn, places the call in an execution environment (e.g., a virtual machine instance or container), such that the called task is executed in the environment. In addition, at (8), the manager 120 returns to the frontend 120 load information for the manager 140, which may be represented as a number of current executions associated with a given arc divided by a number of available environments for the arc managed by the worker manager 140. As discussed below, load information may thereafter be utilized by the frontend 120 to handle subsequent requests to the arc.
While the interactions of
Thus, in accordance with embodiments of the present disclosure, the interactions of
With reference to
The routine 500 begins at block 502, where the frontend 120 obtains a call to execute a task. As discussed above, the call may be transmitted by a user device 102 (e.g., as an API call), from another external device, generated at the frontend 120, etc.
At block 504, the frontend 120 determines whether the call is to a task with reserved capacity. In one embodiment, the frontend 120 may maintain information identifying an arc (e.g., within a table of such information) for each task with reserved concurrency. The frontend 120 may therefore determine whether the call is to a task with reserved capacity by inspecting such information. In another embodiment, each frontend 120 may maintain other information identifying tasks with reserved capacity, and may create information identifying the arc for the task in response to a first call for execution of the task. As discussed above, an arc for reserved capacity may illustratively be created with a static length, set based on the level of reserved capacity. An arc for non-reserved capacity (e.g., an account arc) may be scaled at the frontend 120 based on a load of that arc.
If the called task is not associated with reserved capacity, the routine 500 proceeds to block 506, where the frontend 120 increments a counter associated with a general (e.g., account) throttle limit. Illustratively, the frontend 120 may transmit a call to a counting service (e.g., the service 170) indicating an additional concurrent execution associated with the throttle limit. The routine 120 then proceeds to block 508, where the frontend 120 selects a worker manager 140 from a general (e.g., account) arc. As discussed above, selection of a manager 140 may include identifying managers 140 within the arc based on a hash value of an identifier for the arc (e.g., an account identifier) and based on an arc length, and then selecting among the managers 140 based on a load-balancing mechanism, such as round robin selection. As noted above, the frontend 120 may periodically (e.g., at every 5 minutes) increase or decrease a throttle limit and a length of the arc based on the number of concurrent executions associated with the throttle limit.
If the called task is associated with reserved capacity, the routine 500 proceeds to block 510, where the frontend 120 increments a counter associated with a reserved capacity throttle limit. Illustratively, the frontend 120 may transmit a call to a counting service (e.g., the service 170) indicating an additional concurrent execution associated with the throttle limit. The routine 120 then proceeds to block 512, where the frontend 120 selects a worker manager 140 from a reserved capacity arc. As discussed above, selection of a manager 140 may include identifying managers 140 within the arc based on a hash value of an identifier for the arc (e.g., an account identifier) and based on an arc length (e.g., established based on the reserved capacity), and then selecting among the managers 140 based on a load-balancing mechanism, such as round robin selection.
While not shown in
In either instance, the routine 500 then proceeds to block 514, where the frontend 120 requests from the selected manager 140 an execution environment in which to execute the task. In the case of requesting an environment from the reserved arc, it is expected that the manager 140 has reserved a number of pre-warmed environments for the task, and thus returns such a pre-warmed environment. In this case of requesting an environment from the general arc, a pre-warmed environment may be available to the manager 140 (e.g., as previously managed by the manager 140 or as maintained in a warming pool 130A), in which case the pre-warmed environment is returned. In the case that no pre-warmed environment is available (e.g., at the manager 140 or within a warming pool 130A), the manager 140 may provision a new environment for the task. While not shown in
At block 516, after obtaining an execution environment from a worker manager 140, the frontend 120 causes the called task to be executed within the environment. The routine 500 then ends at block 518. Thus, by execution of the routine 500, the frontend 120 is enabled to route calls to execute both tasks associated with reserved capacity and tasks not associated with reserved capacity, in a manner that enables load balancing among managers 140 while still providing distributing calls in a manner that increases the probability that a pre-warmed environment for the call can be located.
As discussed above, in some instances a frontend 120 may receive calls to a function associated with reserved capacity, with at least some of those calls exceeding the throttle limit for reserved capacity. Those calls that exceed the throttle limit for reserved capacity are generally referred to herein as “spillover” calls. In some instances, a frontend 120 can be configured to route such spillover calls utilizing a more general arc (e.g., an account arc) than an arc associated with reserved capacity. However, as discussed above, in some instances routing spillover calls utilizing a more general arc may result in cold starts even while concurrency of a task's executions remains steady. For example, reducing a level of reserved capacity (or eliminating reserved capacity entirely) while executions remain steady may result in some new calls (representing a portion of the steady execution rate) becoming spillover calls. If such calls are routed to a more general arc, those calls may incur cold start delays, and additionally cause the system 110 to incur greater resource usage, as managers 140 associated with the reserved capacity arc may still be managing environments previously used to support the reserved capacity arc.
To detect that the call exceeds the reserved capacity of the task, the frontend 120 transmits, at (2), an instruction to the counting service 170 to increment a counter associated with a throttle limit for reserved capacity of the task. As discussed above, the service 170 may operate to monitor concurrent executions based on inputs from all frontends 120, and may thus maintain a count of concurrent executions across the system 110. In response to the call, the service 170 returns to the frontend 120 a count of the concurrent executions associated with the reserved capacity throttle limit, at (3). The frontend 120 can then, at (4), detect that the count of concurrent executions exceeds the reserved capacity throttle limit.
As discussed above, the frontend 120 can be configured such that when the count of concurrent executions exceeds reserved capacity throttle limit, a more general (e.g., an account) arc for the task is utilized. However, to guard against potential cold starts in instances of steady-state traffic due to a reduction in reserved capacity, the frontend 120 at (5), prior to switching to a more general arc, assesses a load on the reserved capacity arc. In one embodiment, the load on an arc is represented as an average load across all worker managers 140 in the arc, and the load of each worker manager 140 is represented as a number of in-use (e.g., “hot”) environments associated with the arc divided by the total number of environments (e.g., “hot” or “warm”) for the arc managed by the worker manager 140. Thus, in cases where reserved capacity has not changed and spillover calls are received, the load on the arc can be expected to exceed a threshold amount (e.g., 95%, 99%, etc.). Put in other terms, where a user request 500 reserved environments for a given task and transmits (in some sufficiently short period) 501 calls to execute the task, the 501st call to execute the task can be expected to result in both (i) a counter for the reserved capacity exceeding the level of reserved capacity and (ii) the load on the arc nearing 100%.
However, when a level of reserved capacity has changed (e.g., been reduced), it is conceived that while the counter for the reserved capacity throttle limit may exceed the (now reduced) level of reserved capacity, the load on an arc may be significantly less than 100%. This low load can be expected, for example, because while a frontend may immediately recognize a user's reduced level of reserved capacity, each worker manager 140 can be configured to release environments based on their idleness. Accordingly, in instances where a level of reserved capacity has changed but a concurrency rate for the task has remained steady, the arc for the task can be expected to have sufficient capacity to continue to handle that steady rate of concurrent executions. For the purposes of illustration of
Accordingly, at (6), and even though the call exceeds a current reserved capacity throttle limit, the frontend 120 selects a worker manager 140 from the reserved capacity arc, in accordance with the selection mechanisms described above. While not shown in
In addition, at (7), the frontend 120 transmits a request to the selected worker manager 140 for an execution environment in which to execute the task. However, as opposed to other requests for environments sent from the frontend 120 to the worker manager 140, the frontend 120 may at (7) utilize a “no-scale-up” flag for the request. Illustratively, the manager 140 may be generally be configured such that if the manager 140 does not currently manage an appropriate environment for a call, the manager 140 retrieves such an environment from a warming pool manager 130. This adoption of an environment by the worker manager 140 can be referred to as “scaling up” the manager's environments. Because the worker manager 140 was selected from the reserved capacity arc on the expectation that that arc had idle capacity, it can generally be undesirable for that capacity to be increased due to scaling up. Accordingly, by use of a “no-scale-up” flag at interaction (7), the frontend 120 can request an environment from the manager 140 without potentially causing the manager 140 to increase its capacity.
For the purposes of description of
On receiving identifying information of the execution environment 602, the frontend can then, at (10), instruct the environment 602 to execute the called task. The environment 602, at (11), executes the task. Because the interactions of
With reference to
The routine 700 begins at block 702, where the frontend 120 obtains a call to execute a task. Block 702 may be implemented similarly to block 502 of the routine 500. The routine 700 then proceeds to block 706, where the frontend 120 determines whether the call is associated with reserved capacity, and if so, increments a counter associated with either a reserved capacity throttle limit or a general (e.g., account) throttle limit, at blocks 708, and 710, respectively. Blocks 706-710 may be implemented in a manner similar to corresponding blocks within routine 500 (e.g., blocks 504, 510, and 506).
In contrast to the routine 500, the routine 700 includes, subsequent to incrementing a counter for a reserved concurrency throttle limit at block 708, a decision block 712. At block 712, the frontend 120 determines whether a count of concurrent executions associated with the reserved throttle limit excludes that throttle limit. Illustratively, the frontend 120 may obtain, in response to incrementing a counter associated with the throttle limit, a number of concurrent executions associated with throttle limit across the system 110. Implementation of decision block 712 may thus include comparing this number with the total throttle limit.
In the case that the reserved capacity throttle limit is not exceeded, the routine 700 proceeds to block 722, where the frontend 120 selects a worker manager from the reserved capacity arc, and requests an environment from that manager (similarly to blocks 512 and 514 of the routine 500). The routine 700 then proceeds to block 724, where the frontend 120 causes the environment to execute the task (similarly to block 516 of
However, if at block 712 the count of calls associated with the reserved capacity throttle limit does exceed the throttle limit, the routine 700 instead proceeds to block 710, where the call is handled as a spillover call. As such, the frontend 120 increments, at block 710, a counter associated with the general throttle limit. While not shown in
Thereafter, at block 714, the frontend 120 determines whether the load on a reserved capacity arc associated with the called task is below a threshold value, set equal to or near 100% (e.g., 95%, 99%, etc.). Notably, in the routine 700, block 714 is implemented in both cases where a call is associated with current reserved capacity (and that throttle limit of that reserved capacity is exceeded in terms of a concurrent execution count) and where a call is not associated with current reserved capacity. Typically, where a call is not associated with reserved capacity, it is expected that no reserved capacity arc for that task would exist. Where reserved capacity is exceeded, it may typically be expected that the load on the reserved capacity arc would exceed the threshold. In such cases, the routine 700 proceeds to block 718, where the frontend 120 requests an environment from a worker manager 140 selected from the general arc (similarly to blocks 506 and 508 of the routine 500). The routine 700 then proceeds to block 724, where the frontend 120 causes the environment to execute the task (similarly to block 516 of
However, exceptions to these situations may exist, particularly where a task was previously associated with reserved capacity and that capacity is reduced or eliminated. In such an instance, a reserved capacity arc may exist and have a load under the threshold value, despite the fact that a call is not currently associated with reserved capacity (or reserved capacity is exceeded in terms of a concurrent execution count). Accordingly, if block 714 evaluates to true, the routine 700 proceeds to block 716, where the frontend 120 requests an environment from a worker manager 140 selected from the reserved capacity arc (e.g., selected as discussed above with respect to block 512 of
At block 720, the routine 700 varies according to whether the manager 140 selected from the reserved capacity arc had an available environment in response to the request of block 716. If so, the routine 700 proceeds to block 724, where the frontend 120 causes the environment to execute the task (similarly to block 516 of
The routine 700 then proceeds to block 724, where the frontend 120 causes the environment to execute the task (similarly to block 516 of
All of the methods and processes described above may be embodied in, and fully automated via, software code modules executed by one or more computers or processors. The code modules may be stored in any type of non-transitory computer-readable medium or other computer storage device. Some or all of the methods may alternatively be embodied in specialized computer hardware.
Conditional language such as, among others, “can,” “could,” “might” or “may,” unless specifically stated otherwise, are otherwise understood within the context as used in general to present that certain embodiments include, while other embodiments do not include, certain features, elements and/or steps. Thus, such conditional language is not generally intended to imply that features, elements and/or steps are in any way required for one or more embodiments or that one or more embodiments necessarily include logic for deciding, with or without user input or prompting, whether these features, elements and/or steps are included or are to be performed in any particular embodiment.
Disjunctive language such as the phrase “at least one of X, Y or Z,” unless specifically stated otherwise, is otherwise understood with the context as used in general to present that an item, term, etc., may be either X, Y or Z, or any combination thereof (e.g., X, Y and/or Z). Thus, such disjunctive language is not generally intended to, and should not, imply that certain embodiments require at least one of X, at least one of Y or at least one of Z to each be present.
Unless otherwise explicitly stated, articles such as ‘a’ or ‘an’ should generally be interpreted to include one or more described items. Accordingly, phrases such as “a device configured to” are intended to include one or more recited devices. Such one or more recited devices can also be collectively configured to carry out the stated recitations. For example, “a processor configured to carry out recitations A, B and C” can include a first processor configured to carry out recitation A working in conjunction with a second processor configured to carry out recitations B and C.
Any routine descriptions, elements or blocks in the flow diagrams described herein and/or depicted in the attached figures should be understood as potentially representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or elements in the routine. Alternate implementations are included within the scope of the embodiments described herein in which elements or functions may be deleted, or executed out of order from that shown or discussed, including substantially synchronously or in reverse order, depending on the functionality involved as would be understood by those skilled in the art.
It should be emphasized that many variations and modifications may be made to the above-described embodiments, the elements of which are to be understood as being among other acceptable examples. All such modifications and variations are intended to be included herein within the scope of this disclosure and protected by the following claims.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4949254 | Shorter | Aug 1990 | A |
| 5283888 | Dao et al. | Feb 1994 | A |
| 5970488 | Crowe et al. | Oct 1999 | A |
| 6385636 | Suzuki | May 2002 | B1 |
| 6463509 | Teoman et al. | Oct 2002 | B1 |
| 6501736 | Smolik et al. | Dec 2002 | B1 |
| 6523035 | Fleming et al. | Feb 2003 | B1 |
| 6708276 | Yarsa et al. | Mar 2004 | B1 |
| 7036121 | Casabona et al. | Apr 2006 | B1 |
| 7590806 | Harris et al. | Sep 2009 | B2 |
| 7665090 | Tormasov et al. | Feb 2010 | B1 |
| 7707579 | Rodriguez | Apr 2010 | B2 |
| 7730464 | Trowbridge | Jun 2010 | B2 |
| 7774191 | Berkowitz et al. | Aug 2010 | B2 |
| 7823186 | Pouliot | Oct 2010 | B2 |
| 7886021 | Scheifler et al. | Feb 2011 | B2 |
| 8010990 | Ferguson et al. | Aug 2011 | B2 |
| 8024564 | Bassani et al. | Sep 2011 | B2 |
| 8046765 | Cherkasova et al. | Oct 2011 | B2 |
| 8051180 | Mazzaferri et al. | Nov 2011 | B2 |
| 8051266 | DeVal et al. | Nov 2011 | B2 |
| 8065676 | Sahai et al. | Nov 2011 | B1 |
| 8065682 | Baryshnikov et al. | Nov 2011 | B2 |
| 8095931 | Chen et al. | Jan 2012 | B1 |
| 8127284 | Meijer et al. | Feb 2012 | B2 |
| 8146073 | Sinha | Mar 2012 | B2 |
| 8166304 | Murase et al. | Apr 2012 | B2 |
| 8171473 | Lavin | May 2012 | B2 |
| 8209695 | Pruyne et al. | Jun 2012 | B1 |
| 8219987 | Vlaovic et al. | Jul 2012 | B1 |
| 8321554 | Dickinson | Nov 2012 | B2 |
| 8321558 | Sirota et al. | Nov 2012 | B1 |
| 8336079 | Budko et al. | Dec 2012 | B2 |
| 8352608 | Keagy et al. | Jan 2013 | B1 |
| 8387075 | McCann et al. | Feb 2013 | B1 |
| 8429282 | Ahuja | Apr 2013 | B1 |
| 8448165 | Conover | May 2013 | B1 |
| 8490088 | Tang | Jul 2013 | B2 |
| 8555281 | Van Dijk et al. | Oct 2013 | B1 |
| 8566835 | Wang et al. | Oct 2013 | B2 |
| 8613070 | Borzycki et al. | Dec 2013 | B1 |
| 8631130 | Jackson | Jan 2014 | B2 |
| 8677359 | Cavage et al. | Mar 2014 | B1 |
| 8694996 | Cawlfield et al. | Apr 2014 | B2 |
| 8700768 | Benari | Apr 2014 | B2 |
| 8719415 | Sirota et al. | May 2014 | B1 |
| 8725702 | Raman et al. | May 2014 | B1 |
| 8756696 | Miller | Jun 2014 | B1 |
| 8769519 | Leitman et al. | Jul 2014 | B2 |
| 8799236 | Azari et al. | Aug 2014 | B1 |
| 8799879 | Wright et al. | Aug 2014 | B2 |
| 8806468 | Meijer et al. | Aug 2014 | B2 |
| 8819679 | Agarwal et al. | Aug 2014 | B2 |
| 8825863 | Hansson et al. | Sep 2014 | B2 |
| 8825964 | Sopka et al. | Sep 2014 | B1 |
| 8839035 | Dimitrovich et al. | Sep 2014 | B1 |
| 8850432 | Mcgrath et al. | Sep 2014 | B2 |
| 8874952 | Tameshige et al. | Oct 2014 | B2 |
| 8904008 | Calder et al. | Dec 2014 | B2 |
| 8997093 | Dimitrov | Mar 2015 | B2 |
| 9027087 | Ishaya et al. | May 2015 | B2 |
| 9038068 | Engle et al. | May 2015 | B2 |
| 9052935 | Rajaa | Jun 2015 | B1 |
| 9086897 | Oh et al. | Jul 2015 | B2 |
| 9092837 | Bala et al. | Jul 2015 | B2 |
| 9098528 | Wang | Aug 2015 | B2 |
| 9110732 | Forschmiedt et al. | Aug 2015 | B1 |
| 9110770 | Raju et al. | Aug 2015 | B1 |
| 9111037 | Nalis et al. | Aug 2015 | B1 |
| 9112813 | Jackson | Aug 2015 | B2 |
| 9141410 | Leafe et al. | Sep 2015 | B2 |
| 9146764 | Wagner | Sep 2015 | B1 |
| 9152406 | De et al. | Oct 2015 | B2 |
| 9164754 | Pohlack | Oct 2015 | B1 |
| 9183019 | Kruglick | Nov 2015 | B2 |
| 9208007 | Harper et al. | Dec 2015 | B2 |
| 9218190 | Anand et al. | Dec 2015 | B2 |
| 9223561 | Orveillon et al. | Dec 2015 | B2 |
| 9223966 | Satish et al. | Dec 2015 | B1 |
| 9250893 | Blahaerath et al. | Feb 2016 | B2 |
| 9268586 | Voccio et al. | Feb 2016 | B2 |
| 9298633 | Zhao et al. | Mar 2016 | B1 |
| 9317689 | Aissi | Apr 2016 | B2 |
| 9323556 | Wagner | Apr 2016 | B2 |
| 9361145 | Wilson et al. | Jun 2016 | B1 |
| 9413626 | Reque et al. | Aug 2016 | B2 |
| 9436555 | Dornemann et al. | Sep 2016 | B2 |
| 9461996 | Hayton et al. | Oct 2016 | B2 |
| 9471775 | Wagner et al. | Oct 2016 | B1 |
| 9483335 | Wagner et al. | Nov 2016 | B1 |
| 9489227 | Oh et al. | Nov 2016 | B2 |
| 9497136 | Ramarao et al. | Nov 2016 | B1 |
| 9501345 | Lietz et al. | Nov 2016 | B1 |
| 9514037 | Dow et al. | Dec 2016 | B1 |
| 9537788 | Reque et al. | Jan 2017 | B2 |
| 9575798 | Terayama et al. | Feb 2017 | B2 |
| 9588790 | Wagner et al. | Mar 2017 | B1 |
| 9594590 | Hsu | Mar 2017 | B2 |
| 9596350 | Dymshyts et al. | Mar 2017 | B1 |
| 9600312 | Wagner et al. | Mar 2017 | B2 |
| 9628332 | Bruno, Jr. et al. | Apr 2017 | B2 |
| 9635132 | Lin et al. | Apr 2017 | B1 |
| 9652306 | Wagner et al. | May 2017 | B1 |
| 9652617 | Evans et al. | May 2017 | B1 |
| 9654508 | Barton et al. | May 2017 | B2 |
| 9661011 | Van Horenbeeck et al. | May 2017 | B1 |
| 9678773 | Wagner et al. | Jun 2017 | B1 |
| 9678778 | Youseff | Jun 2017 | B1 |
| 9703681 | Taylor et al. | Jul 2017 | B2 |
| 9715402 | Wagner et al. | Jul 2017 | B2 |
| 9727725 | Wagner et al. | Aug 2017 | B2 |
| 9733967 | Wagner et al. | Aug 2017 | B2 |
| 9760387 | Wagner et al. | Sep 2017 | B2 |
| 9767271 | Ghose | Sep 2017 | B2 |
| 9785476 | Wagner et al. | Oct 2017 | B2 |
| 9787779 | Frank et al. | Oct 2017 | B2 |
| 9811363 | Wagner | Nov 2017 | B1 |
| 9811434 | Wagner | Nov 2017 | B1 |
| 9830175 | Wagner | Nov 2017 | B1 |
| 9830193 | Wagner et al. | Nov 2017 | B1 |
| 9830449 | Wagner | Nov 2017 | B1 |
| 9864636 | Patel et al. | Jan 2018 | B1 |
| 9910713 | Wisniewski et al. | Mar 2018 | B2 |
| 9921864 | Singaravelu et al. | Mar 2018 | B2 |
| 9928108 | Wagner et al. | Mar 2018 | B1 |
| 9929916 | Subramanian et al. | Mar 2018 | B1 |
| 9930103 | Thompson | Mar 2018 | B2 |
| 9930133 | Susarla et al. | Mar 2018 | B2 |
| 9952896 | Wagner et al. | Apr 2018 | B2 |
| 9977691 | Marriner et al. | May 2018 | B2 |
| 9979817 | Huang et al. | May 2018 | B2 |
| 10002026 | Wagner | Jun 2018 | B1 |
| 10013267 | Wagner et al. | Jul 2018 | B1 |
| 10042660 | Wagner et al. | Aug 2018 | B2 |
| 10048974 | Wagner et al. | Aug 2018 | B1 |
| 10061613 | Brooker et al. | Aug 2018 | B1 |
| 10067801 | Wagner | Sep 2018 | B1 |
| 10102040 | Marriner et al. | Oct 2018 | B2 |
| 10108443 | Wagner et al. | Oct 2018 | B2 |
| 10139876 | Lu et al. | Nov 2018 | B2 |
| 10140137 | Wagner | Nov 2018 | B2 |
| 10162672 | Wagner et al. | Dec 2018 | B2 |
| 10162688 | Wagner | Dec 2018 | B2 |
| 10203990 | Wagner et al. | Feb 2019 | B2 |
| 10248467 | Wisniewski et al. | Apr 2019 | B2 |
| 10277708 | Wagner et al. | Apr 2019 | B2 |
| 10303492 | Wagner et al. | May 2019 | B1 |
| 10353678 | Wagner | Jul 2019 | B1 |
| 10353746 | Reque et al. | Jul 2019 | B2 |
| 10365985 | Wagner | Jul 2019 | B2 |
| 10387177 | Wagner et al. | Aug 2019 | B2 |
| 10402231 | Marriner et al. | Sep 2019 | B2 |
| 10437629 | Wagner et al. | Oct 2019 | B2 |
| 10528390 | Brooker et al. | Jan 2020 | B2 |
| 10552193 | Wagner et al. | Feb 2020 | B2 |
| 10564946 | Wagner et al. | Feb 2020 | B1 |
| 10572375 | Wagner | Feb 2020 | B1 |
| 10592269 | Wagner et al. | Mar 2020 | B2 |
| 20010044817 | Asano et al. | Nov 2001 | A1 |
| 20020120685 | Srivastava et al. | Aug 2002 | A1 |
| 20020172273 | Baker et al. | Nov 2002 | A1 |
| 20030071842 | King et al. | Apr 2003 | A1 |
| 20030084434 | Ren | May 2003 | A1 |
| 20030229794 | James, II et al. | Dec 2003 | A1 |
| 20040003087 | Chambliss et al. | Jan 2004 | A1 |
| 20040049768 | Matsuyama et al. | Mar 2004 | A1 |
| 20040098154 | McCarthy | May 2004 | A1 |
| 20040158551 | Santosuosso | Aug 2004 | A1 |
| 20040205493 | Simpson et al. | Oct 2004 | A1 |
| 20040249947 | Novaes et al. | Dec 2004 | A1 |
| 20040268358 | Darling et al. | Dec 2004 | A1 |
| 20050027611 | Wharton | Feb 2005 | A1 |
| 20050044301 | Vasilevsky et al. | Feb 2005 | A1 |
| 20050120160 | Plouffe et al. | Jun 2005 | A1 |
| 20050132167 | Longobardi | Jun 2005 | A1 |
| 20050132368 | Sexton et al. | Jun 2005 | A1 |
| 20050149535 | Frey et al. | Jul 2005 | A1 |
| 20050193113 | Kokusho et al. | Sep 2005 | A1 |
| 20050193283 | Reinhardt et al. | Sep 2005 | A1 |
| 20050237948 | Wan et al. | Oct 2005 | A1 |
| 20050257051 | Richard | Nov 2005 | A1 |
| 20060123066 | Jacobs et al. | Jun 2006 | A1 |
| 20060129684 | Datta | Jun 2006 | A1 |
| 20060184669 | Vaidyanathan et al. | Aug 2006 | A1 |
| 20060200668 | Hybre et al. | Sep 2006 | A1 |
| 20060212332 | Jackson | Sep 2006 | A1 |
| 20060242647 | Kimbrel et al. | Oct 2006 | A1 |
| 20060248195 | Toumura et al. | Nov 2006 | A1 |
| 20070094396 | Takano et al. | Apr 2007 | A1 |
| 20070130341 | Ma | Jun 2007 | A1 |
| 20070174419 | O'Connell et al. | Jul 2007 | A1 |
| 20070192082 | Gaos et al. | Aug 2007 | A1 |
| 20070199000 | Shekhel et al. | Aug 2007 | A1 |
| 20070220009 | Morris et al. | Sep 2007 | A1 |
| 20070240160 | Paterson-Jones | Oct 2007 | A1 |
| 20070255604 | Seelig | Nov 2007 | A1 |
| 20080028409 | Cherkasova et al. | Jan 2008 | A1 |
| 20080052401 | Bugenhagen et al. | Feb 2008 | A1 |
| 20080052725 | Stoodley et al. | Feb 2008 | A1 |
| 20080082977 | Araujo et al. | Apr 2008 | A1 |
| 20080104247 | Venkatakrishnan et al. | May 2008 | A1 |
| 20080104608 | Hyser et al. | May 2008 | A1 |
| 20080126110 | Haeberle et al. | May 2008 | A1 |
| 20080126486 | Heist | May 2008 | A1 |
| 20080127125 | Anckaert et al. | May 2008 | A1 |
| 20080147893 | Marripudi et al. | Jun 2008 | A1 |
| 20080189468 | Schmidt et al. | Aug 2008 | A1 |
| 20080195369 | Duyanovich et al. | Aug 2008 | A1 |
| 20080201568 | Quinn et al. | Aug 2008 | A1 |
| 20080201711 | Amir Husain | Aug 2008 | A1 |
| 20080209423 | Hirai | Aug 2008 | A1 |
| 20090006897 | Sarsfield | Jan 2009 | A1 |
| 20090013153 | Hilton | Jan 2009 | A1 |
| 20090025009 | Brunswig et al. | Jan 2009 | A1 |
| 20090055810 | Kondur | Feb 2009 | A1 |
| 20090055829 | Gibson | Feb 2009 | A1 |
| 20090070355 | Cadarette et al. | Mar 2009 | A1 |
| 20090077569 | Appleton et al. | Mar 2009 | A1 |
| 20090125902 | Ghosh et al. | May 2009 | A1 |
| 20090158275 | Wang et al. | Jun 2009 | A1 |
| 20090177860 | Zhu et al. | Jul 2009 | A1 |
| 20090193410 | Arthursson et al. | Jul 2009 | A1 |
| 20090198769 | Keller et al. | Aug 2009 | A1 |
| 20090204960 | Ben-yehuda et al. | Aug 2009 | A1 |
| 20090204964 | Foley et al. | Aug 2009 | A1 |
| 20090222922 | Sidiroglou et al. | Sep 2009 | A1 |
| 20090271472 | Scheifler et al. | Oct 2009 | A1 |
| 20090288084 | Astete et al. | Nov 2009 | A1 |
| 20090300599 | Piotrowski | Dec 2009 | A1 |
| 20100023940 | Iwamatsu et al. | Jan 2010 | A1 |
| 20100031274 | Sim-Tang | Feb 2010 | A1 |
| 20100031325 | Maigne et al. | Feb 2010 | A1 |
| 20100036925 | Haffner | Feb 2010 | A1 |
| 20100058342 | Machida | Mar 2010 | A1 |
| 20100058351 | Yahagi | Mar 2010 | A1 |
| 20100064299 | Kacin et al. | Mar 2010 | A1 |
| 20100070678 | Zhang et al. | Mar 2010 | A1 |
| 20100070725 | Prahlad et al. | Mar 2010 | A1 |
| 20100106926 | Kandasamy et al. | Apr 2010 | A1 |
| 20100114825 | Siddegowda | May 2010 | A1 |
| 20100115098 | De Baer et al. | May 2010 | A1 |
| 20100122343 | Ghosh | May 2010 | A1 |
| 20100131936 | Cheriton | May 2010 | A1 |
| 20100131959 | Spiers et al. | May 2010 | A1 |
| 20100186011 | Magenheimer | Jul 2010 | A1 |
| 20100198972 | Umbehocker | Aug 2010 | A1 |
| 20100199285 | Medovich | Aug 2010 | A1 |
| 20100257116 | Mehta et al. | Oct 2010 | A1 |
| 20100269109 | Cartales | Oct 2010 | A1 |
| 20100312871 | Desantis et al. | Dec 2010 | A1 |
| 20100325727 | Neystadt et al. | Dec 2010 | A1 |
| 20110010722 | Matsuyama | Jan 2011 | A1 |
| 20110029970 | Arasaratnam | Feb 2011 | A1 |
| 20110040812 | Phillips | Feb 2011 | A1 |
| 20110055378 | Ferris et al. | Mar 2011 | A1 |
| 20110055396 | DeHaan | Mar 2011 | A1 |
| 20110055683 | Jiang | Mar 2011 | A1 |
| 20110078679 | Bozek et al. | Mar 2011 | A1 |
| 20110099204 | Thaler | Apr 2011 | A1 |
| 20110099551 | Fahrig et al. | Apr 2011 | A1 |
| 20110131572 | Elyashev et al. | Jun 2011 | A1 |
| 20110134761 | Smith | Jun 2011 | A1 |
| 20110141124 | Halls et al. | Jun 2011 | A1 |
| 20110153727 | Li | Jun 2011 | A1 |
| 20110153838 | Belkine et al. | Jun 2011 | A1 |
| 20110154353 | Theroux et al. | Jun 2011 | A1 |
| 20110179162 | Mayo et al. | Jul 2011 | A1 |
| 20110184993 | Chawla et al. | Jul 2011 | A1 |
| 20110225277 | Freimuth et al. | Sep 2011 | A1 |
| 20110231680 | Padmanabhan et al. | Sep 2011 | A1 |
| 20110247005 | Benedetti et al. | Oct 2011 | A1 |
| 20110265164 | Lucovsky | Oct 2011 | A1 |
| 20110271276 | Ashok et al. | Nov 2011 | A1 |
| 20110276945 | Chasman et al. | Nov 2011 | A1 |
| 20110314465 | Smith et al. | Dec 2011 | A1 |
| 20110321033 | Kelkar et al. | Dec 2011 | A1 |
| 20120011496 | Shimamura | Jan 2012 | A1 |
| 20120016721 | Weinman | Jan 2012 | A1 |
| 20120041970 | Ghosh et al. | Feb 2012 | A1 |
| 20120054744 | Singh et al. | Mar 2012 | A1 |
| 20120072762 | Atchison et al. | Mar 2012 | A1 |
| 20120072914 | Ota | Mar 2012 | A1 |
| 20120079004 | Herman | Mar 2012 | A1 |
| 20120096271 | Ramarathinam et al. | Apr 2012 | A1 |
| 20120096468 | Chakravorty et al. | Apr 2012 | A1 |
| 20120102307 | Wong | Apr 2012 | A1 |
| 20120102333 | Wong | Apr 2012 | A1 |
| 20120102481 | Mani et al. | Apr 2012 | A1 |
| 20120110155 | Adlung et al. | May 2012 | A1 |
| 20120110164 | Frey et al. | May 2012 | A1 |
| 20120110570 | Jacobson et al. | May 2012 | A1 |
| 20120110588 | Bieswanger et al. | May 2012 | A1 |
| 20120131379 | Tameshige et al. | May 2012 | A1 |
| 20120144290 | Goldman et al. | Jun 2012 | A1 |
| 20120192184 | Burckart et al. | Jul 2012 | A1 |
| 20120197795 | Campbell et al. | Aug 2012 | A1 |
| 20120197958 | Nightingale et al. | Aug 2012 | A1 |
| 20120198442 | Kashyap et al. | Aug 2012 | A1 |
| 20120222038 | Katragadda et al. | Aug 2012 | A1 |
| 20120233464 | Miller et al. | Sep 2012 | A1 |
| 20120331113 | Jain et al. | Dec 2012 | A1 |
| 20130014101 | Ballani et al. | Jan 2013 | A1 |
| 20130042234 | DeLuca et al. | Feb 2013 | A1 |
| 20130054804 | Jana et al. | Feb 2013 | A1 |
| 20130054927 | Raj et al. | Feb 2013 | A1 |
| 20130055262 | Lubsey et al. | Feb 2013 | A1 |
| 20130061208 | Tsao et al. | Mar 2013 | A1 |
| 20130067494 | Srour et al. | Mar 2013 | A1 |
| 20130080641 | Lui et al. | Mar 2013 | A1 |
| 20130097601 | Podvratnik et al. | Apr 2013 | A1 |
| 20130111032 | Alapati et al. | May 2013 | A1 |
| 20130111469 | B et al. | May 2013 | A1 |
| 20130124807 | Nielsen et al. | May 2013 | A1 |
| 20130132942 | Wang | May 2013 | A1 |
| 20130139152 | Chang et al. | May 2013 | A1 |
| 20130139166 | Zhang et al. | May 2013 | A1 |
| 20130151648 | Luna | Jun 2013 | A1 |
| 20130152047 | Moorthi et al. | Jun 2013 | A1 |
| 20130179574 | Calder et al. | Jul 2013 | A1 |
| 20130179881 | Calder et al. | Jul 2013 | A1 |
| 20130179894 | Calder et al. | Jul 2013 | A1 |
| 20130179895 | Calder et al. | Jul 2013 | A1 |
| 20130185719 | Kar et al. | Jul 2013 | A1 |
| 20130185729 | Vasic et al. | Jul 2013 | A1 |
| 20130191924 | Tedesco | Jul 2013 | A1 |
| 20130198319 | Shen et al. | Aug 2013 | A1 |
| 20130198743 | Kruglick | Aug 2013 | A1 |
| 20130198748 | Sharp et al. | Aug 2013 | A1 |
| 20130198763 | Kunze et al. | Aug 2013 | A1 |
| 20130205092 | Roy et al. | Aug 2013 | A1 |
| 20130219390 | Lee et al. | Aug 2013 | A1 |
| 20130227097 | Yasuda et al. | Aug 2013 | A1 |
| 20130227534 | Ike et al. | Aug 2013 | A1 |
| 20130227563 | McGrath | Aug 2013 | A1 |
| 20130227641 | White et al. | Aug 2013 | A1 |
| 20130227710 | Barak et al. | Aug 2013 | A1 |
| 20130232480 | Winterfeldt et al. | Sep 2013 | A1 |
| 20130239125 | Iorio | Sep 2013 | A1 |
| 20130262556 | Xu et al. | Oct 2013 | A1 |
| 20130263117 | Konik et al. | Oct 2013 | A1 |
| 20130275376 | Hudlow et al. | Oct 2013 | A1 |
| 20130275958 | Ivanov et al. | Oct 2013 | A1 |
| 20130275969 | Dimitrov | Oct 2013 | A1 |
| 20130275975 | Masuda et al. | Oct 2013 | A1 |
| 20130283176 | Hoole et al. | Oct 2013 | A1 |
| 20130290538 | Gmach et al. | Oct 2013 | A1 |
| 20130291087 | Kailash et al. | Oct 2013 | A1 |
| 20130297964 | Hegdal et al. | Nov 2013 | A1 |
| 20130311650 | Brandwine et al. | Nov 2013 | A1 |
| 20130339950 | Ramarathinam et al. | Dec 2013 | A1 |
| 20130346946 | Pinnix | Dec 2013 | A1 |
| 20130346964 | Nobuoka et al. | Dec 2013 | A1 |
| 20130346987 | Raney et al. | Dec 2013 | A1 |
| 20130346994 | Chen et al. | Dec 2013 | A1 |
| 20130347095 | Barjatiya et al. | Dec 2013 | A1 |
| 20140007097 | Chin et al. | Jan 2014 | A1 |
| 20140019523 | Heymann et al. | Jan 2014 | A1 |
| 20140019735 | Menon et al. | Jan 2014 | A1 |
| 20140019965 | Neuse et al. | Jan 2014 | A1 |
| 20140019966 | Neuse et al. | Jan 2014 | A1 |
| 20140040343 | Nickolov et al. | Feb 2014 | A1 |
| 20140040857 | Trinchini et al. | Feb 2014 | A1 |
| 20140040880 | Brownlow et al. | Feb 2014 | A1 |
| 20140059209 | Alnoor | Feb 2014 | A1 |
| 20140059226 | Messerli et al. | Feb 2014 | A1 |
| 20140059552 | Cunningham et al. | Feb 2014 | A1 |
| 20140068568 | Wisnovsky | Mar 2014 | A1 |
| 20140068611 | McGrath et al. | Mar 2014 | A1 |
| 20140081984 | Sitsky et al. | Mar 2014 | A1 |
| 20140082165 | Marr et al. | Mar 2014 | A1 |
| 20140101649 | Kamble et al. | Apr 2014 | A1 |
| 20140108722 | Lipchuk et al. | Apr 2014 | A1 |
| 20140109087 | Jujare et al. | Apr 2014 | A1 |
| 20140109088 | Dournov et al. | Apr 2014 | A1 |
| 20140129667 | Ozawa | May 2014 | A1 |
| 20140130040 | Lemanski | May 2014 | A1 |
| 20140137110 | Engle et al. | May 2014 | A1 |
| 20140173614 | Konik et al. | Jun 2014 | A1 |
| 20140173616 | Bird et al. | Jun 2014 | A1 |
| 20140180862 | Certain et al. | Jun 2014 | A1 |
| 20140189677 | Curzi et al. | Jul 2014 | A1 |
| 20140201735 | Kannan et al. | Jul 2014 | A1 |
| 20140207912 | Thibeault | Jul 2014 | A1 |
| 20140215073 | Dow et al. | Jul 2014 | A1 |
| 20140245297 | Hackett | Aug 2014 | A1 |
| 20140279581 | Devereaux | Sep 2014 | A1 |
| 20140280325 | Krishnamurthy et al. | Sep 2014 | A1 |
| 20140282615 | Cavage et al. | Sep 2014 | A1 |
| 20140282629 | Gupta et al. | Sep 2014 | A1 |
| 20140283045 | Brandwine et al. | Sep 2014 | A1 |
| 20140289286 | Gusak | Sep 2014 | A1 |
| 20140298295 | Overbeck | Oct 2014 | A1 |
| 20140304698 | Chigurapati et al. | Oct 2014 | A1 |
| 20140304815 | Maeda | Oct 2014 | A1 |
| 20140317617 | O'Donnell | Oct 2014 | A1 |
| 20140344457 | Bruno, Jr. et al. | Nov 2014 | A1 |
| 20140344736 | Ryman et al. | Nov 2014 | A1 |
| 20140380085 | Rash et al. | Dec 2014 | A1 |
| 20150033241 | Jackson et al. | Jan 2015 | A1 |
| 20150039891 | Ignatchenko et al. | Feb 2015 | A1 |
| 20150040229 | Chan et al. | Feb 2015 | A1 |
| 20150046926 | Kenchammana-Hosekote et al. | Feb 2015 | A1 |
| 20150052258 | Johnson et al. | Feb 2015 | A1 |
| 20150058914 | Yadav | Feb 2015 | A1 |
| 20150067830 | Johansson et al. | Mar 2015 | A1 |
| 20150074659 | Madsen et al. | Mar 2015 | A1 |
| 20150081885 | Thomas et al. | Mar 2015 | A1 |
| 20150106805 | Melander et al. | Apr 2015 | A1 |
| 20150120928 | Gummaraju et al. | Apr 2015 | A1 |
| 20150134626 | Theimer et al. | May 2015 | A1 |
| 20150135287 | Medeiros et al. | May 2015 | A1 |
| 20150142952 | Bragstad et al. | May 2015 | A1 |
| 20150143381 | Chin et al. | May 2015 | A1 |
| 20150178110 | Li et al. | Jun 2015 | A1 |
| 20150186129 | Apte et al. | Jul 2015 | A1 |
| 20150188775 | Van Der Walt et al. | Jul 2015 | A1 |
| 20150199218 | Wilson et al. | Jul 2015 | A1 |
| 20150205596 | Hiltegen et al. | Jul 2015 | A1 |
| 20150227598 | Hahn et al. | Aug 2015 | A1 |
| 20150235144 | Gusev et al. | Aug 2015 | A1 |
| 20150242225 | Muller et al. | Aug 2015 | A1 |
| 20150254248 | Burns et al. | Sep 2015 | A1 |
| 20150256621 | Noda et al. | Sep 2015 | A1 |
| 20150261578 | Greden et al. | Sep 2015 | A1 |
| 20150289220 | Kim et al. | Oct 2015 | A1 |
| 20150309923 | Iwata et al. | Oct 2015 | A1 |
| 20150319160 | Ferguson et al. | Nov 2015 | A1 |
| 20150332048 | Mooring et al. | Nov 2015 | A1 |
| 20150350701 | Lemus et al. | Dec 2015 | A1 |
| 20150356294 | Tan et al. | Dec 2015 | A1 |
| 20150363181 | Alberti et al. | Dec 2015 | A1 |
| 20150370560 | Tan et al. | Dec 2015 | A1 |
| 20150371244 | Neuse et al. | Dec 2015 | A1 |
| 20150378762 | Saladi et al. | Dec 2015 | A1 |
| 20150378764 | Sivasubramanian et al. | Dec 2015 | A1 |
| 20150378765 | Singh et al. | Dec 2015 | A1 |
| 20150379167 | Griffith et al. | Dec 2015 | A1 |
| 20160012099 | Tuatini et al. | Jan 2016 | A1 |
| 20160019536 | Ortiz et al. | Jan 2016 | A1 |
| 20160026486 | Abdallah | Jan 2016 | A1 |
| 20160048606 | Rubinstein et al. | Feb 2016 | A1 |
| 20160072727 | Leafe et al. | Mar 2016 | A1 |
| 20160077901 | Roth et al. | Mar 2016 | A1 |
| 20160098285 | Davis et al. | Apr 2016 | A1 |
| 20160100036 | Lo et al. | Apr 2016 | A1 |
| 20160117254 | Susarla et al. | Apr 2016 | A1 |
| 20160124665 | Jain et al. | May 2016 | A1 |
| 20160140180 | Park et al. | May 2016 | A1 |
| 20160191420 | Nagarajan et al. | Jun 2016 | A1 |
| 20160285906 | Fine et al. | Sep 2016 | A1 |
| 20160292016 | Bussard et al. | Oct 2016 | A1 |
| 20160294614 | Searle et al. | Oct 2016 | A1 |
| 20160306613 | Busi et al. | Oct 2016 | A1 |
| 20160350099 | Suparna et al. | Dec 2016 | A1 |
| 20160357536 | Firlik et al. | Dec 2016 | A1 |
| 20160364265 | Cao et al. | Dec 2016 | A1 |
| 20160371127 | Antony et al. | Dec 2016 | A1 |
| 20160371156 | Merriman | Dec 2016 | A1 |
| 20160378449 | Khazanchi et al. | Dec 2016 | A1 |
| 20160378554 | Gummaraju et al. | Dec 2016 | A1 |
| 20170041309 | Ekambaram et al. | Feb 2017 | A1 |
| 20170060621 | Whipple et al. | Mar 2017 | A1 |
| 20170068574 | Cherkasova et al. | Mar 2017 | A1 |
| 20170075749 | Ambichl et al. | Mar 2017 | A1 |
| 20170083381 | Cong et al. | Mar 2017 | A1 |
| 20170085447 | Chen et al. | Mar 2017 | A1 |
| 20170085591 | Ganda et al. | Mar 2017 | A1 |
| 20170093684 | Jayaraman et al. | Mar 2017 | A1 |
| 20170093920 | Ducatel et al. | Mar 2017 | A1 |
| 20170230499 | Mumick et al. | Aug 2017 | A1 |
| 20170272462 | Kraemer et al. | Sep 2017 | A1 |
| 20170286143 | Wagner et al. | Oct 2017 | A1 |
| 20170371724 | Wagner et al. | Dec 2017 | A1 |
| 20180046453 | Nair et al. | Feb 2018 | A1 |
| 20180046482 | Karve et al. | Feb 2018 | A1 |
| 20180060221 | Yim et al. | Mar 2018 | A1 |
| 20180060318 | Yang | Mar 2018 | A1 |
| 20180067841 | Mahimkar | Mar 2018 | A1 |
| 20180121245 | Wagner et al. | May 2018 | A1 |
| 20180143865 | Wagner et al. | May 2018 | A1 |
| 20180275987 | Vandeputte | Sep 2018 | A1 |
| 20180309819 | Thompson | Oct 2018 | A1 |
| 20190072529 | Andrawes et al. | Mar 2019 | A1 |
| 20190102231 | Wagner | Apr 2019 | A1 |
| 20190108058 | Wagner et al. | Apr 2019 | A1 |
| 20190155629 | Wagner et al. | May 2019 | A1 |
| 20190171470 | Wagner | Jun 2019 | A1 |
| 20190196884 | Wagner | Jun 2019 | A1 |
| 20190227849 | Wisniewski et al. | Jul 2019 | A1 |
| 20190384647 | Reque et al. | Dec 2019 | A1 |
| 20200057680 | Marriner et al. | Feb 2020 | A1 |
| Number | Date | Country |
|---|---|---|
| 2663052 | Nov 2013 | EP |
| 2002287974 | Oct 2002 | JP |
| 2006-107599 | Apr 2006 | JP |
| 2007-538323 | Dec 2007 | JP |
| 2010-026562 | Feb 2010 | JP |
| 2011-233146 | Nov 2011 | JP |
| 2011257847 | Dec 2011 | JP |
| 2013-156996 | Aug 2013 | JP |
| 2014-525624 | Sep 2014 | JP |
| 2017-534107 | Nov 2017 | JP |
| 2017-534967 | Nov 2017 | JP |
| 2018-503896 | Feb 2018 | JP |
| 2018-512087 | May 2018 | JP |
| 2018-536213 | Dec 2018 | JP |
| WO 2008114454 | Sep 2008 | WO |
| WO 2009137567 | Nov 2009 | WO |
| WO 2012050772 | Apr 2012 | WO |
| WO 2013106257 | Jul 2013 | WO |
| WO 2015078394 | Jun 2015 | WO |
| WO 2015108539 | Jul 2015 | WO |
| WO 2016053950 | Apr 2016 | WO |
| WO 2016053968 | Apr 2016 | WO |
| WO 2016053973 | Apr 2016 | WO |
| WO 2016090292 | Jun 2016 | WO |
| WO 2016126731 | Aug 2016 | WO |
| WO 2016164633 | Oct 2016 | WO |
| WO 2016164638 | Oct 2016 | WO |
| WO 2017059248 | Apr 2017 | WO |
| WO 2017112526 | Jun 2017 | WO |
| WO 2017172440 | Oct 2017 | WO |
| Entry |
|---|
| Anonymous: “Docker run reference”, Dec. 7, 2015, XP055350246, Retrieved from the Internet: URL:https://web.archive.org/web/20151207111702/https:/docs.docker.com/engine/reference/run/ [retrieved on Feb. 28, 2017]. |
| Adapter Pattern, Wikipedia, https://en.wikipedia.org/w/index.php?title=Adapter_pattern&oldid=654971255, [retrieved May 26, 2016], 6 pages. |
| Amazon, “AWS Lambda: Developer Guide”, Retrieved from the Internet, Jun. 26, 2016, URL: http://docs.aws.amazon.com/lambda/ latest/dg/lambda-dg.pdf, 346 pages. |
| Amazon, “AWS Lambda: Developer Guide”, Retrieved from the Internet, 2019, URL : http://docs.aws.amazon.com/lambda/ latest/dg/lambda-dg.pdf, 521 pages. |
| Balazinska et al., Moirae: History-Enhanced Monitoring, Published: 2007, 12 pages. |
| Ben-Yehuda et al., “Deconstructing Amazon EC2 Spot Instance Pricing”, ACM Transactions on Economics and Computation 1.3, 2013, 15 pages. |
| Bhadani et al., Performance evaluation of web servers using central load balancing policy over virtual machines on cloud, Jan. 2010, 4 pages. |
| CodeChef ADMIN discussion web page, retrieved from https://discuss.codechef.com/t/what-are-the-memory-limit-and-stack-size-on-codechef/14159, 2019. |
| CodeChef IDE web page, Code, Compile & Run, retrieved from https://www.codechef.com/ide, 2019. |
| Czajkowski, G., and L. Daynes, Multitasking Without Compromise: A Virtual Machine Evolution 47(4a):60-73, ACM SIGPLAN Notices—Supplemental Issue, Apr. 2012. |
| Das et al., Adaptive Stream Processing using Dynamic Batch Sizing, 2014, 13 pages. |
| Deis, Container, 2014, 1 page. |
| Dombrowski, M., et al., Dynamic Monitor Allocation in the Java Virtual Machine, JTRES '13, Oct. 9-11, 2013, pp. 30-37. |
| Dynamic HTML, Wikipedia page from date Mar. 27, 2015, retrieved using the WayBackMachine, from https://web.archive.org/web/20150327215418/https://en.wikipedia.org/wiki/Dynamic_HTML, 2015, 6 pages. |
| Espadas, J., et al., A Tenant-Based Resource Allocation Model for Scaling Software-as-a-Service Applications Over Cloud Computing Infrastructures, Future Generation Computer Systems, vol. 29, pp. 273-286, 2013. |
| Han et al., Lightweight Resource Scaling for Cloud Applications, 2012, 8 pages. |
| Hoffman, Auto scaling your website with Amazon Web Services (AWS)—Part 2, Cardinalpath, Sep. 2015, 15 pages. |
| http://discuss.codechef.com discussion web page from date Nov. 11, 2012, retrieved using the WayBackMachine, from https://web.archive.org/web/20121111040051/http://discuss.codechef.com/questions/2881 /why-are-simple-java-programs-using-up-so-much-space, 2012. |
| https://www.codechef.com code error help page from Jan. 2014, retrieved from https://www.codechef.com/JAN14/status/ERROR,va123, 2014. |
| http://www.codechef.com/ide web page from date Apr. 5, 2015, retrieved using the WayBackMachine, from https://web.archive.org/web/20150405045518/http://www.codechef.com/ide, 2015. |
| Kamga et al., Extended scheduler for efficient frequency scaling in virtualized systems, Jul. 2012, 8 pages. |
| Kazempour et al., AASH: an asymmetry-aware scheduler for hypervisors, Jul. 2010, 12 pages. |
| Kraft et al., 10 performance prediction in consolidated virtualized environments, Mar. 2011, 12 pages. |
| Krsul et al., “VMPlants: Providing and Managing Virtual Machine Execution Environments for Grid Computing”, Supercomputing, 2004. Proceedings of the ACM/IEEESC 2004 Conference Pittsburgh, PA, XP010780332, Nov. 6-12, 2004, 12 pages. |
| Meng et al., Efficient resource provisioning in compute clouds via VM multiplexing, Jun. 2010, 10 pages. |
| Merkel, “Docker: Lightweight Linux Containers for Consistent Development and Deployment”, Linux Journal, vol. 2014 Issue 239, Mar. 2014, XP055171140, 16 pages. |
| Monteil, Coupling profile and historical methods to predict execution time of parallel applications. Parallel and Cloud Computing, 2013, <hal-01228236, pp. 81-89. |
| Nakajima, J., et al., Optimizing Virtual Machines Using Hybrid Virtualization, SAC '11, Mar. 21-25, 2011, TaiChung, Taiwan, pp. 573-578. |
| Qian, H., and D. Medhi, et al., Estimating Optimal Cost of Allocating Virtualized Resources With Dynamic Demand, ITC 2011, Sep. 2011, pp. 320-321. |
| Sakamoto, et al. “Platform for Web Services using Proxy Server”; Research Report from Information Processing Society, Mar. 22, 2002, vol. 2002, No. 31. |
| Shim (computing), Wikipedia, https://en.wikipedia.org/w/index.php?title+Shim_(computing)&oldid+654971528, [retrieved on May 26, 2016], 2 pages. |
| Stack Overflow, Creating a database connection pool, 2009, 4 pages. |
| Tan et al., Provisioning for large scale cloud computing services, Jun. 2012, 2 pages. |
| Vaghani, S.B., Virtual Machine File System, ACM SIGOPS Operating Systems Review 44(4):57-70, Dec. 2010. |
| Vaquero, L., et al., Dynamically Scaling Applications in the cloud, ACM SIGCOMM Computer Communication Review 41(1):45-52, Jan. 2011. |
| Wang et al., “Improving utilization through dynamic VM resource allocation in hybrid cloud environment”, Parallel and Distributed V Systems (ICPADS), IEEE, 2014. Retrieved on Feb. 14, 2019, Retrieved from the internet: URL<https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7097814, 8 pages. |
| Wikipedia “API” pages from date Apr. 7, 2015, retrieved using the WayBackMachine from https://web.archive.org/web/20150407191158/https://en .wikipedia.org/wiki/Application_programming_interface. |
| Wikipedia List_of_HTTP status_codes web page, retrieved from https://en.wikipedia.org/wiki/List_of_HTTP status_codes, 2019. |
| Wikipedia Recursion web page from date Mar. 26, 2015, retrieved using the WayBackMachine, from https://web.archive.org/web/20150326230100/https://en.wikipedia.org/wiki/Recursion_(computer _science), 2015. |
| Wikipedia subroutine web page, retrieved from https://en.wikipedia.org/wiki/Subroutine, 2019. |
| Wu et al., HC-Midware: A Middleware to Enable High Performance Communication System Simulation in Heterogeneous Cloud, Association for Computing Machinery, Oct. 20-22, 2017, 10 pages. |
| Yamasaki et al. “Model-based resource selection for efficient virtual cluster deployment”, Virtualization Technology in Distributed Computing, ACM, Nov. 2007, pp. 1-7. |
| Yue et al., AC 2012-4107: Using Amazon EC2 in Computer and Network Security Lab Exercises: Design, Results, and Analysis, 2012, American Society for Engineering Education 2012. |
| Zheng, C., and D. Thain, Integrating Containers into Workflows: A Case Study Using Makeflow, Work Queue, and Docker, VTDC '15, Jun. 15, 2015, Portland, Oregon, pp. 31-38. |
| International Search Report and Written Opinion in PCT/US2015/052810 dated Dec. 17, 2015. |
| International Preliminary Report on Patentability in PCT/US2015/052810 dated Apr. 4, 2017. |
| Extended Search Report in European Application No. 15846932.0 dated May 3, 2018. |
| International Search Report and Written Opinion in PCT/US2015/052838 dated Dec. 18, 2015. |
| International Preliminary Report on Patentability in PCT/US2015/052838 dated Apr. 4, 2017. |
| Extended Search Report in European Application No. 15847202.7 dated Sep. 9, 2018. |
| International Search Report and Written Opinion in PCT/US2015/052833 dated Jan. 13, 2016. |
| International Preliminary Report on Patentability in PCT/US2015/052833 dated Apr. 4, 2017. |
| Extended Search Report in European Application No. 15846542.7 dated Aug. 27, 2018. |
| International Search Report and Written Opinion in PCT/US2015/064071 dated Mar. 16, 2016. |
| International Preliminary Report on Patentability in PCT/US2015/064071 dated Jun. 6, 2017. |
| International Search Report and Written Opinion in PCT/US2016/016211 dated Apr. 13, 2016. |
| International Preliminary Report on Patentability in PCT/US2016/016211 dated Aug. 17, 2017. |
| International Search Report and Written Opinion in PCT/US2016/026514 dated Jun. 8, 2016. |
| International Preliminary Report on Patentability in PCT/US2016/026514 dated Oct. 10, 2017. |
| International Search Report and Written Opinion in PCT/US2016/026520 dated Jul. 5, 2016. |
| International Preliminary Report on Patentability in PCT/US2016/026520 dated Oct. 10, 2017. |
| International Search Report and Written Opinion in PCT/US2016/054774 dated Dec. 16, 2016. |
| International Preliminary Report on Patentability in PCT/US2016/054774 dated Apr. 3, 2018. |
| International Search Report and Written Opinion in PCT/US2016/066997 dated Mar. 20, 2017. |
| International Preliminary Report on Patentability in PCT/US2016/066997 dated Jun. 26, 2018. |
| International Search Report and Written Opinion in PCT/US/2017/023564 dated Jun. 6, 2017. |
| International Preliminary Report on Patentability in PCT/US/2017/023564 dated Oct. 2, 2018. |
| International Search Report and Written Opinion in PCT/US2017/040054 dated Sep. 21, 2017. |
| International Preliminary Report on Patentability in PCT/US2017/040054 dated Jan. 1, 2019. |
| International Search Report and Written Opinion in PCT/US2017/039514 dated Oct. 10, 2017. |
| International Preliminary Report on Patentability in PCT/US2017/039514 dated Jan. 1, 2019. |
| Extended European Search Report in application No. 17776325.7 dated Oct. 23, 2019. |
| Office Action in European Application No. 17743108.7 dated Jan. 14, 2020. |
| U.S. Appl. No. 16/698,845, Serverless Call Distribution to Implement Spillover While Avoiding Cold Starts, filed Nov. 27, 2019. |
