Method and system for facilitating a converged computation and storage node in a distributed storage system

Abstract

A printed circuit board comprises: a network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller. The memory controller comprises: a fabric controller component configured to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; and a media controller component configured to manage access relating to data stored in a volatile memory media. The FPGA is configured to perform computations relating to data stored via the non-volatile-media controller. The heterogeneous processor is configured to perform computation tasks relating to data stored via the memory controller. The printed circuit board is configured to be plugged in to a rack with a plurality of other plugged-in circuit boards.

Description

BACKGROUND

Field

This disclosure is generally related to the field of data storage. More specifically, this disclosure is related to a method and system for facilitating a converged computation and storage node in a distributed storage system.

Related Art

Today, various storage systems are being used to store and access the ever-increasing amount of digital content. Some technological fields or areas may require analysis of a large amount of data, e.g., big data analysis. In such fields, a distributed storage system or distributed cluster may divide computation-heavy tasks into multiple partitions or sub-tasks, where the sub-tasks may be performed in parallel by multiple nodes in the distributed cluster. A single node in such a distributed cluster can include both computation and storage capability. A single node can be a conventional server in a distributed cluster, and can use a central processing unit (CPU) as a central hub to manage traffic among devices or components in the node or server. Because certain computation workloads may not be friendly or optimal for a CPU, various heterogeneous processors can be used as Peripheral Component Interconnect Express (PCIe) devices to handle the main computation tasks. Examples of heterogeneous processors include a general processing unit (GPU) and a field programmable gate array (FPGA). The heterogeneous processors can generally be used in a conventional server as a PCIe add-in card (AIC). By shifting the main computation tasks from the CPU to the heterogeneous processors, a distributed system can deploy the CPU with a decreased cost and power consumption.

However, the conventional server or node architecture can involve multiple layers of memory copy and bus protocols, which can result in a high consumption of resources and power as well as a sub-optimal operational efficiency. Furthermore, despite the decreased cost of CPU usage in conventional servers, the cost of using CPU cores for management instead of computation tasks still remains high.

SUMMARY

In one embodiment, a printed circuit board comprises: a network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller. The memory controller comprises: a fabric controller component configured to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; and a media controller component configured to manage access relating to data stored in a volatile memory media. The FPGA is configured to perform computations relating to data stored via the non-volatile-media controller. The heterogeneous processor is configured to perform computation tasks relating to data stored via the memory controller.

In some embodiments, the fabric controller is further configured to manage data received and transmitted via an internal bus protocol to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller.

In some embodiments, the printed circuit board further comprises: a power module; a monitoring and operating module; and at least one network interface configured to receive data from another node or computing device and further configured to transmit data to the network controller.

In some embodiments, the printed circuit board is immersed in a liquid cooling medium.

In some embodiments, the network controller is an Ethernet controller, the non-volatile-media controller is a NAND controller, and the memory controller is a DRAM controller.

In some embodiments, the network controller further comprises: at least one static random-access memory (SRAM) or embedded dynamic random-access memory (DRAM); an Ethernet interface; at least one ARM core; a bus switch; and a bus root complex or an endpoint.

In some embodiments, the printed circuit board is configured to be plugged into a rack with a plurality of other plugged-in circuit boards. The printed circuit board comprises a node, the other plugged-in circuit boards comprise nodes, and each node is connected to one of a plurality of switches in the rack.

In some embodiments, a switch is an Ethernet switch or an access switch.

In some embodiments, the rack is immersed with the plugged-in plurality of nodes in a liquid cooling tank. A defective node of the immersed rack is identified. The defective node is removed without affecting operation of the rack or a remainder of the nodes in the rack. The defective node is replaced or repaired to obtain a new node. The new node is plugged into the rack by immersing the new node in the liquid cooling tank at a location previously occupied by the defective node.

Another system provides a method for facilitating operation of a storage system. During operation, the system receives, by a network controller of a device, a request to write first data to a non-volatile memory of the device, wherein the device comprises: the network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller. The memory controller comprises: a fabric controller component configured to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; and a media controller component configured to manage access relating to data stored in a volatile memory media. The system performs, by the heterogeneous processor, computation tasks relating to the first data and data stored via the memory controller. The system performs, by the FPGA, computations relating to the first data and data stored via the non-volatile-media controller. The system writes, by the non-volatile-media controller, the first data to a non-volatile media of the device.

BRIEF DESCRIPTION OF THE FIGURES

FIG. 1A illustrates an architecture of an exemplary node in a storage system, in accordance with the prior art.

FIG. 1B illustrates an exemplary environment for a distributed cluster with resource pooling, in accordance with the prior art.

FIG. 2 illustrates a board as a converged computation and storage node in a storage system, in accordance with an embodiment of the present application.

FIG. 3 illustrates an exemplary Ethernet controller, in accordance with an embodiment of the present application.

FIG. 4A illustrates a design of an exemplary rack with converged computation and storage nodes, in accordance with an embodiment of the present application.

FIG. 4B illustrates the exemplary rack of FIG. 4A in a liquid immersion cooling tank, in accordance with an embodiment of the present application.

FIG. 5 illustrates an exemplary environment for accessing memory in a storage system, in accordance with the prior art.

FIG. 6 illustrates an exemplary environment for accessing memory in a converged computation and storage node, in accordance with an embodiment of the present application.

FIG. 7A presents a flowchart illustrating a method for facilitating operation of a storage system, in accordance with an embodiment of the present application.

FIG. 7B presents a flowchart illustrating a method for facilitating operation of a storage system, in accordance with an embodiment of the present application.

FIG. 8 illustrates an exemplary computer system that facilitates operation of a storage system, in accordance with an embodiment of the present application.

FIG. 9 illustrates an exemplary apparatus that facilitates operation of a storage system, in accordance with an embodiment of the present application.

In the figures, like reference numerals refer to the same figure elements.

DETAILED DESCRIPTION

The following description is presented to enable any person skilled in the art to make and use the embodiments, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the present disclosure. Thus, the embodiments described herein are not limited to the embodiments shown, but are to be accorded the widest scope consistent with the principles and features disclosed herein.

Overview

The embodiments described herein facilitate a distributed storage system with converged computation and storage nodes without using a general CPU, which can improve the efficiency and performance of the storage system.

As described above, in technological fields requiring analysis of large amounts of data (such as big data analysis), a distributed storage system or distributed cluster may divide computation-heavy tasks into multiple partitions or sub-tasks, where the sub-tasks may be performed in parallel by multiple nodes in the distributed cluster. A single node in such a distributed cluster can include both computation and storage capability. A single node can be a conventional server in a distributed cluster, and can use a CPU as a central hub to manage traffic among devices or components in the node or server. Because certain computation workloads may not be friendly or optimal for a CPU, various heterogeneous processors can be used as PCIe devices to handle the main computation tasks. Examples of heterogeneous processors include a GPU and an FPGA. The heterogeneous processors can generally be used in a conventional server as a PCIe AIC. By shifting the main computation tasks from the CPU to the heterogeneous processors, a distributed storage system can deploy the CPU with a decreased cost and power consumption.

However, the conventional server or node architecture can involve multiple layers of memory copy and bus protocols, which can result in a high consumption of resources and power as well a sub-optimal operational efficiency. Furthermore, despite the decreased cost of CPU usage in conventional servers, the cost of using CPU cores for management instead of computation tasks still remains high. An exemplary conventional server is described below in relation to FIGS. 1A and 1B.

The embodiments described herein addresses these challenges by providing a converged computation and storage node which includes a network controller and a memory controller which is directly accessible via the network controller and other components of the node. The node can be a single printed circuit board with multiple components or integrated circuits (ICs). The node can include: the network controller, which can be an Ethernet controller; the memory controller, which can be a DRAM controller (or a “standalone DRAM controller”); a heterogeneous processor; a storage field-programmable gate array (FPGA); and a non-volatile media controller, which can be a NAND controller. The storage FPGA can be configured to perform computations relating to data stored via the NAND controller, and the heterogeneous processor can be configured to perform computation tasks. The NAND controller can be configured to manage high-capacity NAND flash. The Ethernet controller can be configured to communicate with other nodes and extend the data processing and storage functions within an internal bus. Thus, the node can provide both computation and storage functionality, and can be referred to as “a converged computation and storage node.”

In this converged computation and storage node, the standalone DRAM controller can be used and accessed directly by the heterogeneous processor, the storage FPGA, the NAND controller, and the Ethernet controller through a peer-to-peer direct memory access. This converged node can thus eliminate the need for a middle layer between an initiator and a target of the bus, which can result in a decrease in the amount of overhead traditionally associated with copying data to and from memory, as described below in relation to FIGS. 1A, 1B, and 2. Furthermore, the converged node does not need to include a general CPU, which can result in further improvements in performance, cost, and efficiency.

Thus, by placing both computation and storage functionality in a single node (e.g., a printed circuit board), the system can eliminate the need to use a CPU, which can result in an improvement in performance and a reduction in power consumption through use of the heterogeneous processors, which can provide the computation power. In the single node, the system can further provide direct access to the node's memory controller via several other node components, including a heterogeneous processor, a storage FPGA, and a NAND controller. These improvements can result in a more efficient distributed storage system.

A “storage system infrastructure,” “storage infrastructure,” or “storage system” refers to the overall set of hardware and software components used to facilitate storage for a system. A storage system can include multiple clusters of storage servers and other servers. A “storage server” refers to a computing device which can include multiple storage devices or storage drives. A “storage device” or a “storage drive” refers to a device or a drive with a non-volatile memory which can provide persistent storage of data, e.g., a solid state drive (SSD), a hard disk drive (HDD), or a flash-based storage device.

A “computing device” refers to any server, device, node, entity, drive, or any other entity which can provide any computing capabilities.

A “node” refers to a device or printed circuit board which can be used in a distributed storage system. A “converged computation and storage node” refers to a node which includes functionality for both computation and storage, and can include the components, modules, unit, or ICs as described herein.

A “network controller” refers to a controller which can be configured to manage, handle, or otherwise process traffic received from and transmitted via, e.g., an Ethernet interface. A network controller can include the Ethernet interface, SRAM/embedded DRAM (eDRAM), a bus switch, a bus root complex, and ARM cores which provide conversion between Ethernet and an internal bus. A network controller can also communicate with or be coupled to a “memory controller” (see term below). An exemplary network controller is described below in relation to FIG. 3.

A “heterogeneous processor” refers to a processor which can be configured to perform computation tasks, including high-cost tasks requiring high-performance, e.g., related to artificial intelligence and big data analysis.

A “non-volatile-media controller” refers to a controller, component, unit, module, IC, software, firmware, or hardware component which can be configured to manage, process, store, or other handle data stored in or to be stored in a non-volatile memory of a system. An example of a non-volatile media is NAND flash, and an example of a non-volatile-media controller is a NAND controller.

A “storage FPGA” refers to an FPGA which can be configured to perform computations relating to data stored or to be stored, e.g., via a non-volatile media controller of the same node or PCB.

A “memory controller” refers to a controller which can be configured to manage, process, store, or otherwise handle data stored in or to be stored in a volatile memory of a system. An example of a memory controller is a DRAM controller. In this disclosure, a memory controller can include a “fabric controller” and a “media controller.” The fabric controller can be used to manage data received from and transmitted via an internal bus protocol to communicate with other components or devices, including, but not limited to, a network controller, a storage FPGA, a heterogeneous processor, and a non-volatile-media controller. The media controller can be configured to handle the access of data stored in the non-volatile media and to provide reliability assurance of the stored data based on the characteristics of the non-volatile media.

Architecture of Exemplary Node and Environment in the Prior Art

FIG. 1A illustrates an architecture of an exemplary node 100 in a storage system, in accordance with the prior art. Node 100 can include: CPU cores 102 with associated dual in-line memory modules (DIMMs) 104, 106, 108, and 110; a network interface card 122; a heterogeneous processing card 124; and at least one solid state drive (SSD) 126. Node 100 can use CPU cores 102 as a central hub to manage all traffic among devices and components in the node, where the nodes are connect connected with a high-speed Ethernet to form a distributed storage system (or a distributed cluster, as in FIG. 1B).

FIG. 1B illustrates an exemplary environment 130 for a distributed cluster with resource pooling, in accordance with the prior art. Environment 130 can include: a data center Ethernet fabric 132; a plurality of nodes, including a node 140 and a node 150; and master nodes 160. Each node can include a compute module and a storage module. The compute modules across multiple nodes or all nodes of a cluster can form a compute pool, and the storage modules across multiple nodes or all nodes of a cluster can form a storage pool. For example, node 140 can include a compute module 142 and a storage module 144, and node 150 can include a compute module 152 and a storage module 154. Compute modules 142 and 152 can form a compute pool 170, while storage modules 144 and 154 can form a storage pool 172.

At the same time, due to the increasing demand to improve the efficiency of data processing (e.g., in fields such as big data analysis), heterogeneous processors can generally be used in a conventional server as a PCIe AIC. By shifting the main computation tasks from the CPU to the heterogeneous processors, the CPU can be deployed with a decreased cost and low power consumption. However, the conventional server or node architecture can involve multiple layers of memory copy and bus protocols, which can result in high consumption of resources and power, and a sub-optimal operational efficiency. Furthermore, despite the decreased cost of CPU usage in conventional servers, the cost of using CPU cores for management instead of computation tasks still remains high.

In addition to the cost of the CPU, the conventional server is limited by several other constraints. First, CPU sources are limited. Only a few vendors fabricate or manufacture CPUs, which can result in frequent challenges in the supply chain. Second, power utilization can be difficult to control in the conventional server. Because the CPU is no longer performing the main computation tasks (at least in certain compute scenarios), the high power consumption can result in a non-trivial constraints on the deployment of the servers and the optimization of the total cost of operation (TCO). It may be sub-optimal to spend major resources (in terms of cost, components, performance, TCO, etc.) on what may be considered a secondary module (i.e., the CPU). Third, the performance of the storage system can face a bottleneck due to the usage of the CPU and the conventional node and cluster architecture. Moreover, the memory bandwidth, memory copy, and protocol overhead may all act as further constraints on the optimization of the performance of the storage system.

Thus, the conventional server which uses the CPU as the central hub can be limited by several constraints which can limit the optimization of a distributed storage system.

Exemplary Converged Computation and Storage Node

FIG. 2 illustrates a board 200 as a converged computation and storage node in a storage system, in accordance with an embodiment of the present application. Board 200 can be a printed circuit board (PCB), and can include: an Ethernet controller 110 accessed via an Ethernet interface 112 and an Ethernet interface 114; a DRAM controller 120, which is coupled to DRAMs 122, 124, and 126; a heterogeneous processor 130; a storage FPGA 140; a NAND controller 150, which is coupled to NANDs 151-159; a power module 160; an optional fan 162; and a monitoring/operating module 164. Ethernet controller 110 can provide the root complex of a bus which connects DRAM controller 120, heterogeneous processor 130, storage FPGA 140, and NAND controller 150. DRAM controller 120 (a “memory controller”) can be directly used by Ethernet controller 110 (a network controller), heterogeneous processor 130, storage FPGA 140, and NAND controller 150 (a non-volatile-media controller). DRAM controller 120 is accessible via the other components through a peer-to-peer direct memory access. This “converged node” can thus eliminate the need for a middle layer between an initiator and a target of the bus, which can result in a decrease in the amount of overhead traditionally associated with copying data to and from memory.

NAND controller 150 can be configured to manage the high-capacity NAND flash. Storage FPGA 140 can be configured with customized functions to perform computations or computation tasks relating to data stored via NAND controller 150. Heterogeneous processor 130 can be configured to perform computations relating to data stored via DRAM controller 120. These computations can include main computation tasks relating to, e.g., artificial intelligence and big data analysis. Ethernet controller 110 can be configured to communicate with other nodes and to extend the processing and storage of data using the internal bus. Monitoring/operating module 164 can be configured to perform monitoring and operating functions relating to operation of the node, including receiving, processing, accessing, handling, directing, and storing data.

Thus, by providing computation and storage functionality in the above described manner in converged nodes without requiring a central CPU in each node, the embodiments described herein provide converged nodes in a distributed storage system which can improve the optimization and efficiency of the distributed storage system.

Exemplary Network Controller

FIG. 3 illustrates an exemplary Ethernet controller 300, in accordance with an embodiment of the present application. Ethernet controller 300 can include: an SRAM/embedded DRAM (eDRAM) 302; an Ethernet interface 304; Advanced RISC Machines (ARM) cores 306 and 308; a bus switch 310; a bus root complex or endpoint 312; and an SRAM/embedded DRAM (eDRAM) 314. ARM cores 306 and 308 can provide the conversion between the Ethernet and the internal bus (e.g., Ethernet interface 304 and bus switch 310). SRAMs/embedded DRAMs (eDRAMSs) 302 and 314 can provide an internal data buffer for low-latency access.

Exemplary Rack and Immersion in Liquid Cooling Tank

FIG. 4A illustrates a design of an exemplary rack 400 with converged computation and storage nodes, in accordance with an embodiment of the present application. Rack 400 can include a plurality of converged nodes (e.g., 402, 404, 406, 408, 410, and 412) and at least one Ethernet switch (e.g., 422 and 424). Each converged node can be connected or coupled to an Ethernet switch, as depicted by the six lines between each converged node and Ethernet switches 422 and 424. As discussed above, a converged node (as in FIG. 2) can integrate or include multiple ICs on a single printed circuit board. This can result in a straightforward handling of both the mechanical design and the thermal dissipation of the converged nodes and an associated rack, e.g., in a data center. As a result of the mechanical design, because each node is a single PCB which can be plugged into the rack and coupled to a given Ethernet switch, no separate chassis or other module covering or fitting or container is required in order to insert or include the node in the rack. Furthermore, the volume or size of a node as a single PCB is smaller than the volume or size of a conventional server, which can result in occupying less space in the rack and increasing the overall density of nodes in the rack.

In a conventional server rack, the converged storage nodes of FIG. 4A are connected to Ethernet switches 422 and 424 in the middle of the rack. Given the increased density of converged computation and storages nodes in rack 400, the air cooling required to properly handle the thermal dissipation of the converged nodes would still require equipping fans in the converged nodes. One solution is to utilize the mechanical design of the converged storage nodes and place or immerse rack 400 in a liquid cooling medium, such as a liquid immersion cooling tank, as described below in relation to FIG. 4B.

FIG. 4B illustrates exemplary rack 400 of FIG. 4A in a liquid immersion cooling tank 420, in accordance with an embodiment of the present application. Each converged node of rack 400 can be a PCB with multiple electronic devices or ICs, and each single PCB can be placed in its entirety in liquid immersion cooling tank 420. Rack 400 can include converged nodes (402-412) and Ethernet switches (422 and 424) and can also include power cords and Ethernet cables. The cords and cables may or may not be immersed in liquid immersion cooling tank 420. The cooling liquid of liquid immersion cooling tank 420 may not be conductive. The system (or a system administrator, operator, or other user) can identify a defective node or PCB of immersed rack 400, and can remove the defective node or PCB without affecting the operation of rack 400 or the remainder of the nodes or PCBs in rack 400. The system can replace or repair the defective node to obtain a new node, and can plug the new node into the rack by immersing the new nodes in the liquid tank at a location previously occupied by the defective node, or in a different available location in immersed rack 400.

Exemplary Environment for Accessing Memory in the Prior Art vs. in a Converged Computation and Storage Node

FIG. 5 illustrates an exemplary environment 500 for accessing memory in a storage system, in accordance with the prior art. Environment 500 can include: a CPU system on chip (SOC) 510; an SSD 540; a NIC 542; heterogeneous processors 544; and a memory module 530 with DRAMs 532-538. CPU SOC 510 can include: CPU cores 512; an L1 cache 514; an L2 cache 516; an L3 cache 518; a memory controller 520; and a media controller 522. Memory controller 520 communicates with L3 cache 518, where data moves from CPU cores 512 to L3 cache by passing through both of L1 cache 514 and L2 cache 516. Memory controller 520 can also communicate with media controller 522, which in turn can communicate with or otherwise access DRAMs 532-538 of memory module 530. CPU SOC 510 can use multiple PCIe lanes to communicate with SSD 540, NIC 542, and heterogeneous processors 544 (via a communication 562).

In prior art environment 500, memory module 530 is depicted as a passive data bucket because both memory controller 520 and media controller 522 are located on CPU SOC 510. Media controller 512 is designed as a slave of CPU SOC 510, which causes all DRAM usage to travel or go through CPU SOC 510. Using CPU SOC 510 as a central hub like this can result in a significant overhead in the performance of the distributed storage system.

FIG. 6 illustrates an exemplary environment 600 for accessing memory in a converged computation and storage node, in accordance with an embodiment of the present application. Environment 600 can include: an Ethernet controller 610 with an SRAM/eDRAM 612; a storage FPGA 620 with an SRAM/eDRAM 622; a heterogeneous processor 630 which includes heterogeneous processor cores 634 with an SRAM/eDRAM 632; a NAND controller 640 with an SRAM/eDRAM 642; and a memory controller 650 which includes a fabric controller 652 and a media controller 654. NAND controller 640 can communicate with or be coupled to NAND 644, and memory controller 650 (via a media controller 654) can communicate with or be coupled to DRAMs 660-666 and a storage class memory (SCM) 668.

Memory controller 650 can include two components or modules: fabric controller 652 and media controller 654. Fabric controller 652 can be configured to communicate with Ethernet controller 610, heterogeneous processor 630, storage FPGA 620, and NAND controller 640. Media controller 654 can be configured to communicate with volatile memory (e.g., DRAMs 660-666) and other types of memory (e.g., SCM 668). Each of Ethernet controller 610, heterogeneous processor 630, storage FPGA 620, and NAND controller 640 can directly access memory controller 650, and can thus access data stored in or to be stored in DRAMs 660-666 and SCM 668.

Thus, the embodiments described herein provide a distributed storage system with converged computation and storage nodes which communicate via an Ethernet connection. These converged nodes can result in an improvement in the efficiency and performance of the overall distributed storage system by removing the traditional CPU sockets, which can also result in a reduction in both cost and power consumption in scenarios in which a heterogeneous processor can provide computation power. The described memory controller can be shared by the heterogeneous processor, the storage FPGA, the non-volatile-media controller (the NAND controller), and the network controller (the Ethernet controller), but the system can allocate and recycle the memory capacity and space individually without sharing in order to ensure coherence.

Method for Facilitating Operation of a Storage System

FIG. 7A presents a flowchart 700 illustrating a method for facilitating operation of a storage system, in accordance with an embodiment of the present application. During operation, the system receives, by a network controller of a device, a request to write first data to a non-volatile memory of the device, wherein the device comprises: the network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller (operation 702). The memory controller comprises: a fabric controller component configured to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; and a media controller component configured to manage access relating to data stored in a volatile memory media. The system performs, by the heterogeneous processor, computation tasks relating to the first data and data stored via the memory controller (operation 704). The system performs, by the FPGA, computations relating to the first data and data stored via the non-volatile-media controller (operation 706). The system writes, by the non-volatile-media controller, the first data to a non-volatile media of the device (operation 708). The device can be plugged in to a rack with a plurality of other devices.

As described above, the device can be a printed circuit board, and the rack can be immersed into a liquid cooling tank. FIG. 7B presents a flowchart 720 illustrating a method for facilitating operation of a storage system, in accordance with an embodiment of the present application. During operation, a user or an administrative agent can immerse the rack with the plugged-in plurality of nodes in a liquid cooling tank (operation 722). The system can identify a defective node of the immersed rack (operation 724). The user or administrative agent can remove the defective node without affecting operation of the rack or a remainder of the nodes in the rack (operation 726). The system or user or other system can replace or repair the defective node to obtain a new node (operation 728). The system or user or administrative agent can plug the new node into the rack by immersing the new node in the liquid cooling tank at a location previously occupied by the defective node (operation 730).

Exemplary Computer System and Apparatus

FIG. 8 illustrates an exemplary computer system 800 that facilitates operation of a storage system, in accordance with an embodiment of the present application. Computer system 800 includes a processor 802, a volatile memory 806, and a storage device 808. In some embodiments, computer system 800 can include a controller 804 (indicated by the dashed lines). Volatile memory 806 can include, e.g., random access memory (RAM), that serves as a managed memory, and can be used to store one or more memory pools. Storage device 808 can include persistent storage which can be managed or accessed via processor 802 (or controller 804). Furthermore, computer system 800 can be coupled to peripheral input/output (I/O) user devices 810, e.g., a display device 811, a keyboard 812, and a pointing device 814. Storage device 808 can store an operating system 816, a content-processing system 818, and data 836.

Content-processing system 818 can include instructions, which when executed by computer system 800, can cause computer system 800 or processor 802 to perform methods and/or processes described in this disclosure. Specifically, content-processing system 818 can include instructions for receiving and transmitting data packets, including data to be read or written and an input/output (I/O) request (e.g., a read request or a write request) (communication module 820).

Content-processing system 818 can further include instructions for receiving, by a network controller of a device, a request to write first data to a non-volatile memory of the device, wherein the device comprises: the network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller (communication module 820 and network-controlling module 822). The memory controller comprises: a fabric controller component configured to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; and a media controller component configured to manage access relating to data stored in a volatile memory media (memory-controlling module 824). Content-processing system 818 can include instructions for performing, by the heterogeneous processor, computation tasks relating to the first data and data stored via the memory controller (first computation-performing module 826). Content-processing system 818 can include instructions for performing, by the FPGA, computations relating to the first data and data stored via the non-volatile-media controller (second computation-performing module 828). Content-processing system 818 can include instructions for writing, by the non-volatile-media controller, the first data to a non-volatile media of the device (NAND-managing module 832).

Content-processing system 818 can include instructions for identifying a defective node of an immersed rack (defective node-managing module 830), where the rack is immersed with the plugged-in plurality of nodes into a liquid cooling tank. Content-processing system 818 can include instructions for removing the defective node without affecting operation of the rack or a remainder of the nodes in the rack, replacing or repairing the defective node to obtain a new node, and plugging the new node into the rack by immersing the new node into the liquid cooling tank in a location previously occupied by the defective node (actions which can be performed by a node-repairing module (not shown) or by a user or administrative agent.

Data 834 can include any data that is required as input or generated as output by the methods and/or processes described in this disclosure. Specifically, data 834 can store at least: data; a request; a read request; a write request; an input/output (I/O) request; data or metadata associated with a read request, a write request, or an I/O request; an indicator or identifier of a printed circuit board; an indicator or identifier of an interface or other component, circuit, IC, module, or unit on a printed circuit board; an indicator or identifier of a network controller, a processor, a heterogeneous processor, a storage FPGA, a memory controller, or a non-volatile memory controller; an indicator or identifier of a power module, a monitoring and operating module, and a network interface; an indicator or identifier of components of a network controller, including an SRAM or embedded DRAM, an Ethernet interface, at least one ARM core, a bus switch, and a bus root complex or an endpoint; a node or switch identifier; an indication or identifier of a defective node, a new node, or a rack; and any information or data related to, traveling through, stored in, or accessed from a printed circuit board or any component of a printed circuit board as described herein.

FIG. 9 illustrates an exemplary apparatus 900 that facilitates operation of a storage system, in accordance with an embodiment of the present application. Apparatus 900 can comprise a plurality of units or apparatuses which may communicate with one another via a wired, wireless, quantum light, or electrical communication channel. Apparatus 900 may be realized using one or more integrated circuits, and may include fewer or more units or apparatuses than those shown in FIG. 9. Furthermore, apparatus 900 may be integrated in a computer system, or realized as a separate device or devices capable of communicating with other computer systems and/or devices.

Apparatus 900 can comprise modules or units 902-914 which are configured to perform functions or operations similar to modules 820-832 of computer system 800 of FIG. 8, including: a communication unit 902; a network-controlling unit 904; a memory-controlling unit 906; a first computation-performing unit 908; a second computation-performing unit 910; a defective node-managing unit 912; and a NAND-managing unit 914.

The data structures and code described in this detailed description are typically stored on a computer-readable storage medium, which may be any device or medium that can store code and/or data for use by a computer system. The computer-readable storage medium includes, but is not limited to, volatile memory, non-volatile memory, magnetic and optical storage devices such as disk drives, magnetic tape, CDs (compact discs), DVDs (digital versatile discs or digital video discs), or other media capable of storing computer-readable media now known or later developed.

The methods and processes described in the detailed description section can be embodied as code and/or data, which can be stored in a computer-readable storage medium as described above. When a computer system reads and executes the code and/or data stored on the computer-readable storage medium, the computer system performs the methods and processes embodied as data structures and code and stored within the computer-readable storage medium.

Furthermore, the methods and processes described above can be included in hardware modules. For example, the hardware modules can include, but are not limited to, application-specific integrated circuit (ASIC) chips, field-programmable gate arrays (FPGAs), and other programmable-logic devices now known or later developed. When the hardware modules are activated, the hardware modules perform the methods and processes included within the hardware modules.

The foregoing embodiments described herein have been presented for purposes of illustration and description only. They are not intended to be exhaustive or to limit the embodiments described herein to the forms disclosed. Accordingly, many modifications and variations will be apparent to practitioners skilled in the art. Additionally, the above disclosure is not intended to limit the embodiments described herein. The scope of the embodiments described herein is defined by the appended claims.

Claims

1. A printed circuit board (PCB), comprising: a network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller,wherein the PCB does not include a central processing unit;wherein the memory controller comprises: a fabric controller component configured to communicate via a peer-to-peer direct memory access with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller;a media controller component configured to manage access relating to data stored in a volatile memory media,wherein the FPGA is configured to perform computations relating to data stored via the non-volatile-media controller, andwherein the heterogeneous processor is configured to perform computation tasks relating to data stored via the memory controller,wherein the PCB is one of a plurality of PCBs with no respective separate coverings which are each plugged into a rack and coupled to one of a plurality of switches in the rack,wherein the rack, including the plugged-in PCBs with no separate coverings and the switches, is immersed in a liquid cooling tank,wherein a defective plugged-in PCB with no separate covering of the immersed rack is removed without affecting operation of the rack or a remainder of the plugged-in PCBs in the rack, andwherein a new PCB is plugged into the rack by immersing the new PCB with no separate covering in the liquid cooling tank at a first location previously occupied by the defective plugged-in PCB.

2. The printed circuit board of claim 1, wherein the fabric controller is further configured to manage data received and transmitted via an internal bus protocol to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller.

3. The printed circuit board of claim 1, further comprising: a power module;a monitoring and operating module; andat least one network interface configured to receive data from another node or computing device and further configured to transmit data to the network controller.

4. The printed circuit board of claim 1, wherein the printed circuit board is immersed in a liquid cooling medium.

5. The printed circuit board of claim 1, wherein the network controller is an Ethernet controller,wherein the non-volatile-media controller is a NAND controller, andwherein the memory controller is a DRAM controller.

6. The printed circuit board of claim 1, wherein the network controller further comprises: at least one static random-access memory (SRAM) or embedded dynamic random-access memory (DRAM);an Ethernet interface;at least one ARM core;a bus switch; anda bus root complex or an endpoint.

7. The printed circuit board of claim 1, wherein the printed circuit board comprises a node,wherein the other plugged-in circuit boards comprise nodes, andwherein each node is connected to one of the plurality of switches in the rack.

8. The printed circuit board of claim 7, wherein a switch is an Ethernet switch or an access switch.

9. The printed circuit board of claim 7, wherein a defective node of the immersed rack is identified;wherein the defective node is removed without affecting operation of the rack or a remainder of the nodes in the rack;wherein the defective node is replaced or repaired to obtain a new node; andwherein the new node is plugged into the rack by immersing the new node in the liquid cooling tank at a location previously occupied by the defective node.

10. A computer-implemented method, comprising: receiving, by a network controller of a device, a request to write first data to a non-volatile memory of the device, wherein the device comprises: the network controller; a memory controller; a heterogeneous processor; a field-programmable gate array (FPGA); and a non-volatile-media controller, andwherein the memory controller comprises: a fabric controller component configured to communicate via a peer-to-peer direct memory access with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; anda media controller component configured to manage access relating to data stored in a volatile memory media;performing, by the heterogeneous processor, computation tasks relating to the first data and data stored via the memory controller;performing, by the FPGA, computations relating to the first data and data stored via the non-volatile-media controller; andwriting, by the non-volatile-media controller, the first data to a non-volatile media of the device,wherein the device comprises a printed circuit board (PCB) which is one of a plurality of printed circuit boards (PCBs) with no respective coverings which are each plugged into a rack and coupled to one of a plurality of switches in the rack, wherein the PCB does not include a central processing unit,immersing the rack, including the plugged-in plurality of PCBs with no separate coverings and the switches, in a liquid cooling tankremoving a defective plugged-in PCB with no separate covering of the immersed rack without affecting operation of the rack or a remainder of the plugged-in PCBs in the rack; andplugging a new PCB into the rack by immersing the new PCB with no separate covering in the liquid cooling tank at a first location previously occupied by the defective plugged-in PCB.

11. The method of claim 10, wherein the fabric controller is further configured to manage data received and transmitted via an internal bus protocol to communicate with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller.

12. The method of claim 10, wherein the device further comprises: a power module;a monitoring and operating module; andat least one network interface configured to receive data from another node or computing device and further configured to transmit data to the network controller.

13. The method of claim 10, wherein the device is immersed in a liquid cooling medium.

14. The method of claim 10, wherein the network controller is an Ethernet controller,wherein the non-volatile-media controller is a NAND controller, andwherein the memory controller is a DRAM controller.

15. The method of claim 10, wherein the network controller further comprises: at least one static random-access memory (SRAM) or embedded dynamic random-access memory (DRAM);an Ethernet interface;at least one ARM core;a bus switch; anda bus root complex or an endpoint.

16. The method of claim 10, wherein the device is plugged into a rack with a plurality of other devices,wherein the device comprises a node,wherein the other plugged-in devices are printed circuit boards which comprise nodes, andwherein each node is connected to one of a plurality of switches in the rack.

17. The method of claim 16, wherein a switch is an Ethernet switch or an access switch.

18. The method of claim 16, further comprising: identifying a defective node of the immersed rack;removing the defective node without affecting operation of the rack or a remainder of the nodes in the rack;replacing or repairing the defective node to obtain a new node; andplugging the new node into the rack by immersing the new node in the liquid cooling tank at a location previously occupied by the defective node.

19. A memory controller, wherein the memory controller resides on a printed circuit board (PCB),wherein a network controller, a heterogeneous processor, a field-programmable gate array (FPGA), and a non-volatile-media controller reside on the printed circuit board (PCB), wherein the PCB does not include a central processing unit,wherein the memory controller comprises: a fabric controller component configured to manage data received and transmitted via an internal bus protocol to communicate directly with the network controller, the heterogeneous processor, the FPGA, and the non-volatile-media controller; anda media controller component configured to manage access relating to data stored in a volatile memory media,wherein the FPGA is configured to perform computations relating to data stored via the non-volatile-media controller,wherein the heterogeneous processor is configured to perform computation tasks relating to data stored via the memory controller,wherein the PCB is configured to be plugged in to a rack with a plurality of other plugged-in circuit boards,wherein each of the plugged-in PCBs has no separate covering and is coupled to one of a plurality of switches in the rack,wherein the rack, including the plugged-in PCBs with no separate coverings and the switches, is immersed in a liquid cooling tank,wherein a defective plugged-in PCB with no separate covering of the immersed rack is removed without affecting operation of the rack or a remainder of the plugged-in PCBs in the rack, andwherein a new PCB is plugged into the rack by immersing the new PCB with no separate covering in the liquid cooling tank at a first location previously occupied by the defective plugged-in PCB.

20. The memory controller of claim 19, wherein the printed circuit board comprises a node,wherein the other plugged-in circuit boards comprise nodes,wherein each node is connected to one of a plurality of switches in the rack, wherein the rack is immersed with the plugged-in plurality of nodes into a liquid cooling tank,wherein a defective node of the immersed rack is identified,wherein the defective node is removed without affecting operation of the rack or a remainder of the nodes in the rack,wherein the defective node is replaced or repaired to obtain a new node, andwherein the new node is plugged into the rack by immersing the new node into the liquid cooling tank in a location previously occupied by the defective node.