Distributed heterogeneous parallel acceleration system

By introducing heterogeneous devices into distributed systems and using MPI framework management, the parallel computing power of distributed systems is improved, the problem of insufficient computing efficiency in the existing technology is solved, and more efficient computing performance is achieved.

CN119088545APending Publication Date: 2024-12-06山东浪潮数据库技术有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411068244.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing distributed computing systems have shortcomings in parallel computing power and efficiency, especially when processing large-scale data, it is difficult to fully utilize the potential of heterogeneous computing devices.

Method used

Design a distributed heterogeneous parallel acceleration system, by mounting heterogeneous devices such as GPUs and FPGAs on the nodes, and using the MPI parallel computing framework to uniformly manage these devices, realize data horizontal segmentation and state synchronization, and optimize computing task allocation and data transmission.

Benefits of technology

The parallel computing power and efficiency of distributed systems have been significantly improved, and the computing performance has been improved through the rational allocation and management of heterogeneous devices, and the problem of degraded data transmission performance has been solved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088545A_ABST
    Figure CN119088545A_ABST
Patent Text Reader

Abstract

The invention provides a distributed heterogeneous parallel acceleration system, which belongs to the technical field of heterogeneous acceleration processing, and comprises a plurality of nodes, heterogeneous devices of different types and different numbers are mounted on each node, and all the mounted heterogeneous devices are logically combined into a heterogeneous device pool through an MPI parallel computing framework. Therefore, the parallel computing capability of the distributed computing system is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of distributed data processing and heterogeneous acceleration processing technology, and in particular to a distributed heterogeneous parallel acceleration system. Background Art

[0002] A distributed computing system is a system consisting of multiple computers that are interconnected through a network and work together to complete computing tasks. These computers can be distributed in different geographical locations, and they share resources and work together to solve complex computing problems. The key features of distributed computing systems include parallel processing, resource sharing, scalability, and fault tolerance.

[0003] When building a distributed computing system, many factors need to be considered, such as network communication, data consistency, load balancing, security, etc. In order to ensure the efficient and stable operation of the system, these factors need to be carefully planned and optimized. With the continuous advancement of technology, distributed computing systems are also constantly developing and evolving. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a distributed heterogeneous parallel acceleration system, which aims to introduce heterogeneous computing devices such as GPU, FPGA, etc. into the distributed system to improve the parallel computing capability of the distributed system.

[0005] The technical solution of the present invention is:

[0006] A distributed heterogeneous parallel acceleration system.

[0007] It contains several nodes, each of which is equipped with heterogeneous devices of different types and numbers. All mounted heterogeneous devices are logically combined into a heterogeneous device pool through the MPI parallel computing framework.

[0008] Furthermore,

[0009] Each distributed node is equipped with a different number of GPUs or FPGAs. All heterogeneous devices under different nodes form a heterogeneous computing pool, which is uniformly managed through the MPI parallel computing framework.

[0010] Going further,

[0011] The MPI heterogeneous computing process queries the current node status, receives system computing tasks and computing data through the data communication interface provided by the distributed system.

[0012] Going further,

[0013] According to the data redundancy characteristics of the distributed architecture, the redundant distributed nodes in each computing task will send half of the computing data to heterogeneous computing devices through data horizontal segmentation when sending computing tasks.

[0014] Going further,

[0015] The nodes synchronize the computation status through MPI messages.

[0016] Going further,

[0017] The amount of computing data within the same node is evenly distributed according to the number of heterogeneous devices.

[0018] Going further,

[0019] Idle nodes receive computation data from neighboring nodes through MPI for parallel computing.

[0020] Going further,

[0021] Each time the neighboring node data is passed forward only one node (rank+1) or backward (rank-1). BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the architecture of the present invention;

[0023] Figure 2 It is a software architecture and data processing schematic diagram of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] The present invention provides a distributed heterogeneous parallel acceleration system, which improves the parallel computing capability of a distributed computing system by introducing heterogeneous devices such as GPU and FPGA.

[0026] Attached Figure 1 The physical structure of the present invention is shown. The distributed system includes multiple nodes. Each node can mount heterogeneous devices of different types and numbers. All mounted heterogeneous devices are logically combined into a heterogeneous device pool through the MPI parallel computing framework.

[0027] The technical solution working process is as follows:

[0028] 1) Each distributed node is equipped with a different number of GPUs or FPGAs. All heterogeneous devices under different nodes form a heterogeneous computing pool, which is uniformly managed through the MPI parallel computing framework.

[0029] 2) The MPI heterogeneous computing process queries the current node status, receives system computing tasks and computing data, etc. through the data communication interface provided by the distributed system;

[0030] 3) Based on the data redundancy characteristics of the distributed architecture, the redundant distributed nodes in each computing task send half of the computing data to heterogeneous computing devices through data horizontal segmentation when sending computing tasks, and the computing status is synchronized between nodes through MPI messages;

[0031] 4) The amount of computing data in the same node is evenly distributed according to the number of heterogeneous devices;

[0032] 5) Idle nodes can receive computational data from neighboring nodes through MPI for parallel computation. Considering the degradation of computational performance caused by data transmission across nodes, the present invention recommends that neighboring node data be transmitted forward (rank+1) or backward (rank-1) to only one node at a time;

[0033] Attached Figure 2 It shows that the functions of the present invention are concentrated in the MPI heterogeneous computing software. The schematic diagram shows the logical level of the software in the architecture. The technical solution provided by the invention needs to open up data communication with the distributed management software, which may bring certain security risks. It can be flexibly achieved through technical means during implementation.

[0034] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A distributed heterogeneous parallel acceleration system, characterized in that: It contains several nodes, each of which is equipped with heterogeneous devices of different types and numbers. All mounted heterogeneous devices are logically combined into a heterogeneous device pool through the MPI parallel computing framework.

2. The system according to claim 1, characterized in that Each distributed node is equipped with a different number of GPUs or FPGAs. All heterogeneous devices under different nodes form a heterogeneous computing pool, which is uniformly managed through the MPI parallel computing framework.

3. The system according to claim 2, characterized in that The MPI heterogeneous computing process queries the current node status, receives system computing tasks and computing data through the data communication interface provided by the distributed system.

4. The system according to claim 3, characterized in that According to the data redundancy characteristics of the distributed architecture, the redundant distributed nodes in each computing task send half of the computing data to heterogeneous computing devices through data horizontal segmentation when sending computing tasks.

5. The system according to claim 4, characterized in that The nodes synchronize the computation status through MPI messages.

6. The system according to claim 4 or 5, characterized in that: The amount of computing data within the same node is evenly distributed according to the number of heterogeneous devices.

7. The system according to claim 6, characterized in that Idle nodes receive computation data from neighboring nodes through MPI for parallel computing.

8. The system according to claim 7, characterized in that Each time the neighboring node data is passed forward only one node (rank+1) or backward (rank-1).