CN112512262B

CN112512262B - A drawer-type high-density FPGA cloud platform chassis

Info

Publication number: CN112512262B
Application number: CN202011478372.7A
Authority: CN
Inventors: 张科; 于磊; 王亚洲; 常轶松; 赵然; 陈明宇
Original assignee: Institute of Computing Technology of CAS
Current assignee: Institute of Computing Technology of CAS
Priority date: 2020-12-15
Filing date: 2020-12-15
Publication date: 2025-08-05
Anticipated expiration: 2040-12-15
Also published as: WO2022126937A1; CN112512262A

Abstract

The present invention proposes a drawer-type, high-density FPGA cloud platform chassis, comprising: a switching module located at the bottom of the chassis, a power supply module located above the switching module, and a drawer structure located above the power supply module; a control board and an FPGA node board are housed within the drawer structure, with the FPGA node board plugging into the control board via a preset interface; the power transmission end of the power supply module is electrically connected to the power input interfaces of the switching module and the control board; the network switching interface of the switching module is connected to the network interface of the FPGA node board for data exchange between the FPGA node boards. This invention significantly increases the deployable density of FPGA node boards within the FPGA cloud platform chassis, reduces wiring costs, assembly and unassembly complexity, and maintenance difficulty within the chassis, and provides users with a comprehensive and convenient development environment through the use of a self-developed control and management system. Real-time monitoring of the chassis and board status is achieved, and preset interfaces are used to reduce manual wiring, thereby improving the reliability of the FPGA cloud platform chassis.

Description

Drawer type high-density FPGA cloud platform case

Technical Field

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

Background

In recent years, because of the advantages of high energy efficiency, parallel computing, multiple times of programming and the like, the FPGA gradually becomes one of application schemes considered in various fields of computers, particularly in the fields of media compression, encryption and decryption, AI, big data processing and the like, and compared with the traditional CPU and GPU, the FPGA scheme can often achieve energy efficiency improvement of several times or even tens of times. Because of this trend, FPGA cloud platforms have grown.

The FPGA chassis is located on the FPGA cloud platform, just like the commercial standard x86 chassis is located on the cloud computing platform. As a core hardware component of the FPGA cloud platform, the FPGA case is mainly based on a standard x86 server and assisted with an FPGA hardware plug-in card in a PCIe interface form on an x86 mainboard inside the case. In a traditional FPGA cloud platform chassis, a power supply, a fan, a motherboard, a service board card and the like are usually placed on the same layer or the same plane. The modules are typically placed back and forth and connected by cables. Such an organization makes the chassis assembly complex. The circuit board card assembly in the case is mostly composed of an x86 server main board (control board card) and a plurality of FPGA node board cards (service board cards), and golden fingers of the PCIe interface of the FPGA node board cards are inserted into the x86 server control board card for data interaction. Under the general condition, the number of the deployable FPGA nodes is not more than 8 according to the size of the case of the FPGA cloud platform case and the number of interfaces of the main board x86 server PCIe, so that the FPGA node resources capable of providing services in a single case are limited. Under the dual limitations of the existing cloud computing framework and the commercial x86 server chassis, if large-scale deployment is carried out, a large number of x86 servers are required to bear the FPGA node board cards, so that the deployment cost is increased, and the utilization rate of the physical space of a machine room is reduced. Therefore, in the case mode of the traditional FPGA cloud platform, the FPGA nodes are difficult to deploy in a large scale and high density.

In addition, the current commercial FPGA cloud platform generally provides an x86 server and FPGA nodes as a resource to cloud users. The user can develop own application software in the x86 server, and can also finish the development of FPGA acceleration logic in the x86 server. This approach would result in wasted resources and increased costs for users who only need FPGA resources.

Thirdly, the existing FPGA node board card is managed and configured through an x86 server main board in the case, a user must first access the x86 server, then start a related process in the x86 server to carry out corresponding management and configuration on the FPGA node board card, and the flexibility is not strong.

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

Aiming at the existing FPGA cloud platform chassis, when the chassis is installed, power supply and data interaction of the board card are connected through cables, each module and the board card are installed separately, the assembly difficulty is high, and the maintenance is not flexible enough.

The user cannot directly carry out relevant configuration and application on the FPGA node board card, and if the user needs to access the FPGA node board card, the FPGA node board card can be realized through an x86 server.

The data interaction between the FPGA node boards in the case cannot be directly performed.

Disclosure of Invention

Aiming at the defects of the existing cloud platform architecture, the invention effectively reduces the deployment and maintenance cost of the FPGA cloud platform chassis, greatly improves the deployment density of the FPGA node board card in the chassis with limited space, and can save more economic cost when carrying out large-scale FPGA node deployment because an x86 server is not used as a management board card of the FPGA node;

the independently developed control management system is used, the FPGA node board card in the chassis can be managed and configured more reasonably and efficiently, and the resource utilization efficiency of the cloud platform is improved.

Aiming at the defects of the prior art, the invention provides a drawer type high-density FPGA cloud platform chassis, which comprises:

Exchange module at the bottom of the case and above the exchange module a drawer structure located above the power module;

A control board card and an FPGA node board card are arranged in the drawer structure, and the FPGA node board card is inserted into the control board card through a preset interface;

The power transmission end of the power supply module is electrically connected with the power input interface of the switching module and the power input interface of the control board card, and the network switching interface of the switching module is connected with the network interface of the FPGA node board card and is used for interacting data between the FPGA node board cards.

The drawer type high-density FPGA cloud platform chassis comprises a drawer structure, wherein a cooling fan is further arranged in the drawer structure, and a handle for assembling and disassembling is arranged at the tail part of the drawer structure.

The drawer type high-density FPGA cloud platform chassis is characterized in that the FPGA node board card is further provided with a PCIe interface, and the FPGA node board card can be connected with an x86 server through the PCIe interface.

The drawer type high-density FPGA cloud platform chassis comprises a power supply module for standby power supply.

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

The drawer type high-density FPGA cloud platform chassis is characterized in that a sliding rail track is arranged on the inner side of the side wall of the chassis and is connected with the drawer structure through the sliding rail track.

The drawer type high-density FPGA cloud platform chassis is characterized in that the FPGA node board card power supply module is connected to the power taking clamp of the control board card through a power supply copper bar.

The drawer type high-density FPGA cloud platform chassis further comprises a constraint device for fixing the FPGA node board card, wherein the constraint device comprises an opening screen plate positioned on one side of the FPGA node board card, a front panel positioned on the other side of the FPGA node board card and a cover plate positioned above the FPGA node board card.

The drawer type high-density FPGA cloud platform chassis comprises a plurality of drawer structures which are arranged on the power supply module side by side.

The drawer type high-density FPGA cloud platform chassis is characterized in that the control board card controls and manages the FPGA node board card through the preset interface and monitors the working state of the board card in real time.

According to the scheme, the invention has the advantage that the deployable density of the FPGA node board card in the case of the FPGA cloud platform is greatly improved. The wiring cost in the case is reduced, the assembly complexity is simplified, and the maintenance difficulty is reduced. By using an independently developed control management system, a more comprehensive and more convenient development environment is provided for users. And the states of the chassis and the board card are monitored in real time, unnecessary wiring is reduced, and the reliability of the FPGA cloud platform chassis is improved.

Drawings

FIG. 1 is a diagram of the overall structure of the present invention;

FIG. 2 is an internal construction view of a single drawer structure of the present invention;

FIG. 3 is a side view of the present invention;

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

Detailed Description

Aiming at the defects of the existing cloud platform architecture, the invention effectively reduces the deployment and maintenance cost of the FPGA cloud platform chassis 1, can greatly improve the deployment density of the FPGA node board card 8 in a limited chassis space by using a stacked layout and double-drawer structure, and provides a convenient, quick and cheaper FPGA resource use environment for users by a cloud platform computing system and an application method thereof (application number 201810532745.0) and a control management system designed in a method, a device and a system for realizing an FPGA server (application number 202010019013.9).

The technical difficulty of the application is how to realize the deployment of the 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, the application comprises the following key points:

The key point 1 is that the integrated drawer structure rail is shown in fig. 4, and when the drawer structure 2 is installed, the drawer structure is firstly downwards placed along the vertical part of the rail 13, and then inwards pushed along the horizontal part of the rail 13. The rails 13 are positioned on two sides of the inner wall of the chassis 1. The drawer structure 2 is internally provided with all working elements (a control board card 6, a high-density business board card, a cooling fan 12 and the like), and can be used as an independent system to work by itself after being externally connected with a power supply;

The key point 2 is that the stacking structure is arranged, the drawer structure 2, the power supply module 3 and the 100G exchange module 4 in the case 1 are arranged up and down, the depth space of the case 1 is saved, the technical effects are that the space is efficiently utilized, and the node density in the case 1 is improved;

The key point 3 is that an independent control management unit is used for controlling and managing the FPGA node board card 8 and monitoring the working state of the board card in real time through the control board card 6;

The key point 4 is that the preset interface 10 is designed, the state information, the control information and the debugging interface of the FPGA node board card 8 can be interacted through the preset interface 10, and the technical effect is that the internal connection of the case 1 is simplified.

In order to make the above features and effects of the present invention more clearly understood, the following specific examples are given with reference to the accompanying drawings.

As shown in fig. 1 and 3, the whole 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, so that the deployment of the FPGA node board card 8 with higher density is realized. The whole chassis 1 is provided with a double drawer structure 2, a power supply module 3 and a 100G exchange module 4 from top to bottom. As shown in fig. 2, the two drawer structures 2 have the same structure and are mainly used for placing high-density service boards and heat dissipation systems thereof, and two handles 5 are arranged at the tail parts for loading and unloading. The power supply module 3 is composed of three sub power supply modules, and is powered by a dual-purpose one-standby mode, wherein two sub power supply modules work, and the other sub power supply module is standby. The 100G exchange module 4 is placed at the bottommost part of the chassis 1 and is used for data interaction between the FPGA node boards 8. The power supply module 3 needs to supply power to the control board card 6 and the FPGA node board card 8, which are both positioned at the upper part of the case 1, and the power supply module is positioned in the middle and is closer to the board card, so that the length of the power supply copper bar 14 can be shortened. The switching module 4 is mainly connected with a network interface of the FPGA node board card 8 and is used for data interaction between the FPGA node board cards 8.

At present, two integrated drawer structures 2 can be placed inside the case 1, each drawer structure 2 forms a small system, independent heat dissipation and management modes are provided, the case 1 only needs to provide a power supply module 3 for the drawer structure 2, the maintenance is convenient, the replacement is simple, the assembly and the disassembly are easy, and the power supply module 3 is connected to a power taking clamp of a control board card 6 of the drawer structure 2 through a power supply copper bar 14, and the concrete form is shown in fig. 4. The FPGA node board card 8 in the drawer structure 2 is provided with a standard PCIe x16 interface besides a preset interface of the FPGA node board card 8, the interface is used for controlling management and data interaction of the board card, and the FPGA node board card 8 can be directly inserted on an x86 server commonly used in the market at present through the interface, so that the compatibility is strong.

The drawer structure 2 is internally integrated with a control board card 6, and the FPGA node board card 8 is inserted on the control board card 6 through a preset interface 10. The preset interface 10 not only provides power for the FPGA node board card 8, but also integrates an ethernet path (which can support gigabit, tera and higher-speed networks), a monitoring management path, a configuration debugging path and the like, so as to realize functions of power supply, monitoring, configuration debugging and the like of the FPGA board card. The control board card 6 can be further provided with PCIe exchange chips and reserved with a plurality of PCIe interfaces for PCIe data exchange between the FPGA node boards. The FPGA node board card can be inserted on a preset interface and a PCIe interface at the same time, manual connection is not needed in the drawer structure 2, and reliability is high. The drawer structure 2 can realize the management and use of the resources of the FPGA node board card 8 only by providing a network interface.

The control board card 6 integrates an intelligent management system, can monitor information such as temperature, power consumption and the like in the case 1 in real time, and can automatically trigger a safety mechanism after exceeding a limit. The intelligent management system can dynamically schedule the resources of the FPGA node board card 8, and provides a safe and reliable use environment for users. Structurally, the control board 6 can also play a role in supporting the FPGA node board 8.

The four directions all have the constraint to fix it about when FPGA node integrated circuit board 8 installs, and the opening otter board 11 position on the left side organic case 1 has the front panel 7 of FPGA node integrated circuit board 8 on the right side, and PCIe interface and the interface 10 of predetermineeing of below all can support, and the integrated circuit board top has apron 9 to be fixed, sound construction.

Each drawer structure 2 is provided with an independent cooling system fan 12, a cooling air channel is smooth, the control board card 6 has no high-performance CPU, the cooling problem of the control board card 6 is not considered too much, and the two-stage fans 12 work together to dissipate heat when the two drawer structures 2 are placed in the case 1, so that the cooling performance is improved.

The invention has the key points that the density of the FPGA node board cards 8 in the limited space in the case is greatly improved through two integrated drawer structures, the drawer structure 2 is provided with a heat dissipation mode and a power supply mode for the FPGA node board cards 8, the independent work of the case 1 (external power supply is needed) can be separated, the control board cards 6 in the drawer structure 2 and the FPGA node board cards 8 are interconnected through self-defined golden fingers (preset interfaces 10), a large number of power supply modules 3 and management network wiring in the traditional case are removed, and the installation and the debugging are convenient.

The invention can deploy 32 high-performance full-height three-quarter-length FPGA node board cards 8 in a standard 5U server case at present.

The chassis 1 can be adjusted into a single drawer or multi-drawer structure 2 according to the actual requirements of the number of the boards or the physical size of the FPGA boards. For example, if the lateral dimensions of the FPGA board are shorter, the drawer structure 2 may be correspondingly shorter. Then a plurality of (greater than or equal to three) drawer structures 2 can be placed in series in a standard server chassis, thereby improving the physical space utilization of the server rack to a greater extent.

Claims

1. A drawer-type high-density FPGA cloud platform chassis, characterized by comprising:

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

The drawer structure is provided with a control board and multiple FPGA node boards, 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 switching module and the control board, and the network switching interface of the switching module is connected to the network interface of the FPGA node board for exchanging data between the FPGA node boards;

A slide rail is provided on the inner side of the side wall of the chassis, which is connected to the drawer structure through the slide rail. The drawer structure is lowered along the vertical part of the slide rail and then pushed inward along the horizontal part of the rail so that the FPGA node board power supply module is connected to the power clip of the control board through the power supply copper bus.

2. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the drawer structure also includes a cooling fan, and a handle for loading and unloading is installed at the tail of the drawer structure.

3. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the FPGA node board also has a PCIe interface, and the FPGA node board can be connected to the x86 server through the PCIe interface.

4. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the power supply module includes a sub-power supply module for backup power supply.

5. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the chassis is a standard 5U server chassis.

6. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that it also includes a restraint device for fixing the FPGA node board card, and the restraint device includes: an open mesh plate located on one side of the FPGA node board card, a front panel located on the other side of the FPGA node board card, and a cover plate located above the FPGA node board card.

7. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the chassis includes a plurality of drawer structures placed side by side on the power supply module.

8. The drawer-type high-density FPGA cloud platform chassis as described in claim 1 is characterized in that the control board controls and manages the FPGA node board through the preset interface and monitors the working status of the board in real time.