WO2022126937A1

WO2022126937A1 - Drawer-type high-density fpga cloud platform case

Info

Publication number: WO2022126937A1
Application number: PCT/CN2021/085808
Authority: WO
Inventors: 张科; 于磊; 王亚洲; 常轶松; 赵然; 陈明宇
Original assignee: 中国科学院计算技术研究所
Priority date: 2020-12-15
Filing date: 2021-04-07
Publication date: 2022-06-23
Also published as: CN112512262A

Abstract

Provided is a drawer-type high-density FPGA cloud platform case, comprising: an exchange module located at the bottom of the case, a power supply module located on the exchange module, and a drawer structure located on the power supply module, wherein a control board card and FPGA node board cards are arranged in the drawer structure, and the FPGA node board cards are inserted into the control board card by means of preset interfaces; a power transmission end of the power supply module is electrically connected to the exchange module and a power input interface of the control board card; and a network exchange interface of the exchange module is connected to network interfaces of the FPGA node board cards and used for carrying out data interaction between the FPGA node board cards. By means of the present invention, the deployment density of the FPGA node board cards in the FPGA cloud platform case is greatly improved, thereby reducing the wiring cost, mounting and dismounting complexity and maintenance difficulty in the case; a comprehensive and convenient development environment is provided for a user by using a control and management system which is independently researched and developed; and the case and the state of the board cards are monitored in real time, and the number of manually connected wires is reduced by means of the preset interfaces, thereby improving the reliability of the FPGA cloud platform case.

Description

A drawer-type high-density FPGA cloud platform chassis

technical field

The invention relates to the field of computer architecture and FPGA (Field Programmable Gate Array) heterogeneous acceleration, in particular to a cloud platform server based on a programmable gate array.

Background technique

In recent years, FPGA has gradually become one of the application solutions considered in various fields of computer due to its advantages of high energy efficiency, parallel computing and multiple programming, especially in media compression, encryption and decryption, AI, big data processing and other fields, FPGA solutions are more traditional CPU and GPU can often achieve several times or even dozens of times of energy efficiency improvement. Due to this trend, the FPGA cloud platform came into being.

The FPGA chassis is to the FPGA cloud platform what the commercial standard x86 chassis is to the cloud computing platform. As the core hardware component of the FPGA cloud platform, the FPGA chassis is mainly based on a standard x86 server, supplemented by an FPGA hardware card in the form of a PCIe interface on the x86 motherboard inside the chassis. Inside the traditional FPGA cloud platform chassis, power supplies, fans, motherboards, and service boards are usually placed on the same layer or on the same plane. The modules are generally placed in front of each other and connected by cables. Such an organizational structure complicates chassis assembly. The circuit board components in the chassis are mostly composed of an x86 server motherboard (control board) and multiple FPGA node boards (service boards). The FPGA node board is inserted into the x86 server control board through the gold finger of the PCIe interface. data interaction. Usually, the FPGA cloud platform chassis can deploy no more than 8 FPGA nodes according to its own chassis size and the number of PCIe interfaces of the motherboard x86 server, so the FPGA node resources that can provide services in a single chassis are limited. Under the dual constraints of the existing cloud computing framework and commercial x86 server chassis, if a large-scale deployment is carried out, a large number of x86 servers are required to carry FPGA node boards, which leads to higher deployment costs and lower utilization of physical space in the computer room. Therefore, in this traditional FPGA cloud platform chassis mode, it is difficult for FPGA nodes to be deployed on a large scale and at high density.

In addition, the current commercial FPGA cloud platforms usually provide x86 servers and FPGA nodes as resources to cloud users. Users can develop their own application software in the x86 server, and can also complete the development of FPGA acceleration logic in the x86 server. This method will result in waste of resources and increased cost for users who only need FPGA resources.

Third, the existing FPGA node boards are managed and configured through the x86 server motherboard in the chassis. Users must first access the x86 server, and then start the relevant processes in the x86 server to manage and configure the FPGA node boards. Configuration, flexibility is not strong.

According to the architecture of the existing cloud platform chassis x86+ service board, high-density FPGA node deployment cannot be performed in a single chassis.

For the existing FPGA cloud platform chassis, when the chassis is installed, the power supply and data interaction of the boards are mostly connected by cables, and the modules and boards are installed separately, which is difficult to assemble and inflexible to maintain.

Users cannot directly configure and apply the FPGA node board. If there is a need to access the FPGA node board, it can only be achieved through an x86 server.

The data interaction between the FPGA node boards inside the chassis cannot be carried out directly.

Invention Disclosure

Aiming at the shortcomings of the existing cloud platform architecture, the invention effectively reduces the cost of FPGA cloud platform chassis deployment and maintenance; in the limited space chassis, the deployment density of FPGA node boards is greatly improved; since x86 servers are not used as FPGA nodes Management boards, which can save more economic costs when deploying large-scale FPGA nodes;

Using the self-developed control management system, the FPGA node boards in the chassis can be managed and configured more reasonably and efficiently, and the resource utilization efficiency of the cloud platform can be improved.

In view of the deficiencies of the prior art, the present invention proposes a drawer-type high-density FPGA cloud platform chassis, which includes:

A switch module located at the bottom of the chassis, a power supply module located above the switch module, and a drawer structure located above the power supply module;

The drawer structure is provided with a control board and an FPGA node board, and the FPGA node board is plugged into the control board through a preset interface;

The power transmission end of the power supply module is electrically connected to the power input interface of the switch module and the control board, and the network switch interface of the switch module is connected to the network interface of the FPGA node board, and is used to exchange the communication between the FPGA node boards. data.

In the drawer-type high-density FPGA cloud platform chassis, the drawer structure further includes a cooling fan, and a handle for loading and unloading is installed at the rear of the drawer structure.

In the drawer-type high-density FPGA cloud platform chassis, the FPGA node board further has a PCIe interface, and the FPGA node board can be connected to the x86 server through the PCIe interface.

In the drawer-type high-density FPGA cloud platform chassis, the power supply module includes an electronic power supply module for backup power supply.

In the drawer-type high-density FPGA cloud platform chassis, the chassis is a standard 5U server chassis.

In the drawer-type high-density FPGA cloud platform chassis, the inner side of the side wall of the chassis is provided with a slide rail, which is connected to the drawer structure through the slide rail.

In the drawer-type high-density FPGA cloud platform chassis, the power supply module of the FPGA node board card is connected to the power take-off clip of the control board card through a power supply copper bar.

The drawer-type high-density FPGA cloud platform chassis further includes a restraint device for fixing the FPGA node board, and the restraint device includes: an opening mesh plate located on one side of the FPGA node board, located on the other side of the FPGA node board. The front panel on one side, the cover above the FPGA node board.

In the drawer-type high-density FPGA cloud platform chassis, the chassis includes a plurality of the drawer structures arranged side by side on the power supply module.

In the drawer-type high-density FPGA cloud platform chassis, the control board controls and manages the FPGA node board through the preset interface and monitors the working state of the board in real time.

It can be seen from the above solutions that the present invention has the advantages that the present invention will greatly increase the deployable density of the FPGA node boards in the FPGA cloud platform chassis. Reduce the wiring cost in the chassis, simplify the assembly complexity, and reduce the maintenance difficulty. By using the self-developed control management system, it provides users with a more comprehensive and convenient development environment. Real-time monitoring of the chassis and board status and reduce unnecessary wiring, improve the reliability of the FPGA cloud platform chassis.

Brief Description of Drawings

Fig. 1 is the overall structure diagram of the present invention;

Fig. 2 is the internal structure diagram of the single drawer structure of the present invention;

Figure 3 is a side view of the present invention;

FIG. 4 is a partial view of the chassis of the present invention.

Best Mode for Carrying Out the Invention

Aiming at the shortcomings of the existing cloud platform architecture, the present invention effectively reduces the deployment and maintenance costs of the FPGA cloud platform chassis 1; the stacking layout and the double-drawer structure can greatly improve the deployment of the FPGA node board 8 in a limited chassis space Density; through the control management system designed in "a cloud platform computing system and its application method" (application number 201810532745.0) and "a method, device and system for implementing an FPGA server" (application number 202010019013.9), more efficient Manage and configure FPGA node boards, providing users with a convenient, fast and cheaper FPGA resource usage environment.

The technical difficulty of the present invention lies in how to realize the deployment of high-density FPGA nodes in the limited cloud platform chassis space, and how to effectively manage, configure and use the high-density FPGA nodes in the chassis. Specifically, this application includes the following key points:

Key point 1: The integrated drawer structure track is shown in Figure 4. When the drawer structure 2 is installed, first put down along the vertical part of the track 13, and then push it inward along the horizontal part of the track 13. The rails 13 are located on both sides of the inner wall of the chassis 1 . The drawer structure 2 has all the working elements (control board 6, high-density business board, cooling fan 12, etc.), and can work as an independent system after external power supply; technical effect: easy to use and fast to assemble;

Key point 2: Stacked structure layout, the drawer structure 2, power supply module 3, and 100G switching module 4 in the chassis 1 are arranged up and down, saving the depth space of the chassis 1; Technical effect: efficient use of space and increase the node density in the chassis 1;

Key point 3: Independent control and management unit, through the control board 6 to complete the control and management of the FPGA node board 8 and real-time monitoring of the working status of the board; technical effect: rational allocation of resources, monitoring board status;

Key point 4: The design of the preset interface 10, the status information, control information and debugging interface of the FPGA node board 8 can be interacted with through the preset interface 10; technical effect: Simplify the internal wiring of the chassis 1.

In order to make the above-mentioned features and effects of the present invention more clearly and comprehensible, embodiments are given below, and detailed descriptions are given below in conjunction with the accompanying drawings.

As shown in FIG. 1 and FIG. 3 , the entire chassis 1 is laid out in a stacked structure, and the depth space of the chassis 1 is reasonably utilized to accommodate two drawer structures 2 , thereby realizing the deployment of higher-density FPGA node boards 8 . The chassis 1 is composed of a double drawer structure 2 , a power supply module 3 and a 100G switching module 4 respectively from top to bottom. As shown in Figure 2, the two drawer structures 2 have the same structure, and are mainly used for placing high-density service boards and their cooling systems, and two handles 5 are installed at the rear for loading and unloading. The power supply module 3 is composed of three sub-power supply modules and adopts a dual-use and one-standby mode for power supply. Two sub-power supply modules work, and the other sub-power supply module is standby. The 100G switching module 4 is placed at the bottom of the chassis 1 for data interaction between the FPGA node boards 8 . The power supply module 3 needs to supply power to the control board 6 and the FPGA node board 8. They are both located on the upper part of the chassis 1. Placing it in the middle will be closer to the board, which can shorten the length of the power supply copper bar 14. The exchange module 4 is mainly connected with the network interface of the FPGA node board 8, and is used for data interaction between the FPGA node boards 8.

At present, two integrated drawer structures 2 can be placed inside the chassis 1. Each drawer structure 2 constitutes a small system with independent heat dissipation and management methods. The chassis 1 only needs to provide the power supply module 3 for the drawer structure 2, which is convenient for maintenance. , the replacement is simple, the assembly and disassembly are easy, the power supply module 3 is connected to the power taking clip of the control board 6 of the drawer structure 2 through the power supply copper bar 14, and the specific form is shown in Figure 4. In addition to its own preset interface, the FPGA node board 8 in the drawer structure 2 also has a standard PCIe x16 interface. This interface is used for control management and data interaction of the board. The FPGA node board 8 can directly use this interface. It is plugged into the common x86 server on the market and has strong compatibility.

The drawer structure 2 integrates a control board 6 inside, and the FPGA node board 8 is inserted into the control board 6 through a preset interface 10 . In addition to providing power for the FPGA node board 8, the preset interface 10 also integrates an Ethernet channel (which can support Gigabit, 10 Gigabit and higher-speed networks), monitoring and management channels, configuration and debugging channels, etc. Card power supply, monitoring, configuration debugging and other functions. The control board 6 may also be additionally equipped with a PCIe switch chip and reserve multiple PCIe interfaces for PCIe data exchange between the FPGA node boards. The FPGA node board can be plugged into the preset interface and the PCIe interface at the same time, and the drawer structure 2 does not need manual wiring, and has high reliability. The drawer structure 2 only needs to provide a network interface to the outside, so as to realize the management and use of the resources of the FPGA node board 8.

The control board 6 integrates an intelligent management system, which can monitor the temperature, power consumption and other information in the chassis 1 in real time, and automatically trigger the safety mechanism when the limit is exceeded. The intelligent management system can dynamically schedule 8 resources of the FPGA node board to provide users with a safe and reliable use environment. Structurally, the control board 6 can also play the role of supporting the FPGA node board 8 .

When the FPGA node board 8 is installed, there are constraints to fix it in four directions. The left side has the opening mesh board 11 on the chassis 1, the right side has the front panel 7 of the FPGA node board 8, and the PCIe interface and The preset interface 10 can be supported, and the cover plate 9 is fixed above the board card, and the structure is firm.

Each drawer structure 2 has its own independent cooling system fan 12, and the cooling air ducts are unobstructed. The control board 6 does not have a high-performance CPU, so there is no need to think too much about the heat dissipation of the control board 6. Two drawer structures 2 are placed in the chassis 1. When the two-stage fans 12 work together to dissipate heat, the heat dissipation performance is improved.

The key point of the present invention is that the density of the FPGA node board 8 in the limited space in the chassis is greatly improved through two integrated drawer structures. The drawer structure 2 itself has a way of dissipating heat and supplying power to the FPGA node board 8, which can be separated from the The chassis 1 works alone (external power supply is required), and the control board 6 in the drawer structure 2 and the FPGA node board 8 are interconnected through a custom gold finger (preset interface 10), which removes a large number of power supply modules 3 inside the traditional chassis and management network wiring for easy installation and commissioning.

At present, the present invention can deploy 32 high-performance FPGA node boards 8 with full height and three-quarter length in a standard 5U server chassis.

The chassis 1 can be adjusted to a single-drawer or multi-drawer structure 2 according to the actual requirements of the number of boards or the physical size of the FPGA boards. For example, if the lateral dimension of the FPGA board is shorter, the drawer structure 2 can also be shortened accordingly. Then, multiple (greater than or equal to three) drawer structures 2 can be placed in succession in a standard server chassis, thereby improving the physical space utilization of the server rack to a greater extent.

Industrial applicability

The invention provides a drawer-type high-density FPGA cloud platform chassis, comprising: a switch module located at the bottom of the chassis, a power supply module located on the switch module, and a drawer structure located on the power supply module; the drawer structure is provided with a control board card and an FPGA node The board, the FPGA node board is plugged into the control board through a preset interface; the power transmission end of the power supply module is electrically connected to the switching module and the power input interface of the control board, and the network switching interface of the switching module is connected to the network of the FPGA node board. The interfaces are connected to exchange data between the FPGA node boards. The present invention will greatly improve the deployable density of the FPGA node boards in the FPGA cloud platform chassis. Reduce wiring costs, loading and unloading complexity and maintenance difficulty within the chassis. By using the self-developed control management system, it provides users with a comprehensive and convenient development environment. Real-time monitoring of the status of the chassis and board cards, reducing the number of manual wiring through the preset interface, and improving the reliability of the FPGA cloud platform chassis.

Claims

A drawer-type high-density FPGA cloud platform chassis, characterized in that it includes:

A switch module located at the bottom of the chassis, a power supply module located above the switch module, and a drawer structure located above the power supply module;

The drawer structure is provided with a control board and an FPGA node board, and the FPGA node board is plugged into the control board through a preset interface;

The power transmission end of the power supply module is electrically connected to the power input interface of the switch module and the control board, and the network switch interface of the switch module is connected to the network interface of the FPGA node board, and is used to exchange the communication between the FPGA node boards. data.
The drawer-type high-density FPGA cloud platform chassis of claim 1, wherein the drawer structure further includes a cooling fan, and a handle for loading and unloading is installed at the rear of the drawer structure.
The drawer-type high-density FPGA cloud platform chassis according to claim 1, wherein the FPGA node board further has a PCIe interface, and the FPGA node board can be connected to the x86 server through the PCIe interface.
The drawer-type high-density FPGA cloud platform chassis of claim 1, wherein the power supply module includes a sub-power supply module for backup power supply.
The drawer-type high-density FPGA cloud platform chassis of claim 1, wherein the chassis is a standard 5U server chassis.
The drawer-type high-density FPGA cloud platform chassis according to claim 1, wherein a slide rail track is provided on the inner side of the side wall of the chassis, and the slide rail track is connected to the drawer structure.
The drawer-type high-density FPGA cloud platform chassis according to claim 1, wherein the power supply module of the FPGA node board card is connected to the power take-off clip of the control board card through a power supply copper bar.
The drawer-type high-density FPGA cloud platform chassis according to claim 1, further comprising a constraining device for fixing the FPGA node board, the constraining device comprising: an opening mesh plate located on one side of the FPGA node board, The front panel on the other side of the FPGA node board, and the cover above the FPGA node board.
The drawer-type high-density FPGA cloud platform chassis of claim 1, wherein the chassis includes a plurality of the drawer structures arranged side by side on the power supply module.
The drawer-type high-density FPGA cloud platform chassis according to claim 1, wherein the control board controls and manages the FPGA node board through the preset interface and monitors the working state of the board in real time.