Comprehensive optimization method and device for many-to-many communication mode, equipment and medium

By acquiring the network topological features of high-performance computers, the implementation of many-to-many communication mode is optimized, and the problem of low many-to-many communication efficiency is solved, the communication efficiency is improved and the performance of parallel programs is improved.

CN119012234APending Publication Date: 2024-11-22BEIJING KAPULA SCI&TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411070641.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The low communication efficiency of many-to-many communication mode in high-performance computers leads to performance bottlenecks in parallel programs, mainly due to limited bandwidth capacity and high latency caused by competition in communication network usage.

Method used

By acquiring the network topological characteristics of high-performance computers, determining the bandwidth capacity and delay of point-to-point communication, determining the optimization implementation method of many-to-many communication based on this information, and using direct transmission or redistribution between processes at the same level to improve communication efficiency.

Benefits of technology

It effectively improves the efficiency of many-to-many communication, reduces communication competition, and improves the performance of parallel programs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119012234A_ABST
    Figure CN119012234A_ABST
Patent Text Reader

Abstract

The invention provides a comprehensive optimization method and device for a many-to-many communication mode, equipment and a medium, and relates to the technical field of communication, and the method comprises the steps: responding to a many-to-many communication operation initiated by a process group of an application program, and obtaining a network topology structure feature of a high-performance computer running the application program; determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics; determining a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each piece of point-to-point communication; and completing the many-to-many communication operation according to the comprehensive optimization communication implementation mode. According to the invention, the communication efficiency of a many-to-many communication mode is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a comprehensive optimization method, device, equipment and medium for a many-to-many communication mode. Background Art

[0002] As the scale of the problems handled by applications increases rapidly, the amount of computation required by applications increases rapidly. Therefore, it is necessary to develop parallel versions of these applications to take advantage of the many processor cores on modern high-performance computers to significantly accelerate the computation process. The development of parallel versions usually requires the parallel running and collaborative computing of programs on multiple processes, so it is necessary to use tools or programming specifications that support inter-process data communication (such as the message passing interface, or MPI).

[0003] Since the network data transmission speed between computing nodes in high-performance computers is much slower than the computing speed of the processor and the memory access speed, the communication overhead becomes the main factor affecting or even determining the parallel efficiency and parallel scalability of applications. The communication overhead usually includes the overhead of establishing a communication link and the overhead of transmitting data. One way to reduce the communication overhead is to increase the communication bandwidth (that is, the size of the communication message per unit time). On the one hand, the bandwidth size for communication between two processes is affected by the size of the communication message, where the larger the message, the higher the bandwidth; on the other hand, the communication bandwidth capacity is determined by the network topology between the two processes, which can be specifically manifested as follows:

[0004] (1) When two processes are in the same computing node, communication between them can be achieved through memory copying, so there can be a high communication bandwidth capacity;

[0005] (2) When two processes are between different computing nodes in the same cabinet, the communication between them only needs to go through the switch in the cabinet, and usually does not compete with the data communication between other computing nodes for the use of the communication network, so a higher communication bandwidth capacity can be achieved;

[0006] (3) When two processes are in different cabinets, the data communication between the two processes needs to compete with the data communication between other processes for the cross-cabinet communication bandwidth, so the communication bandwidth capacity will be relatively limited;

[0007] (4) When there are multiple pairs of processes communicating between two computing nodes, these communications will compete for the communication bandwidth capacity between the two computing nodes.

[0008] Many-to-many is a commonly used collective communication mode, that is, a group of processes cooperate to complete data communication between each other, in which each process sends data to multiple processes and receives data from multiple processes. The all-to-all communication mode, that is, each process sends data to all processes other than itself and each process receives data from all processes other than itself, can be understood as a special many-to-many communication mode. The many-to-many communication mode is prone to serious competition in the use of communication networks, so the communication efficiency is low and it is easy to become a performance bottleneck for parallel programs. Summary of the invention

[0009] In view of this, an embodiment of the present application provides a comprehensive optimization method for a many-to-many communication mode to improve the efficiency of many-to-many communication.

[0010] The embodiment of the present application provides the following technical solution: a comprehensive optimization method for a many-to-many communication mode, the method comprising:

[0011] In response to a many-to-many communication operation initiated by a process group of an application, obtaining network topology characteristics of a high-performance computer running the application;

[0012] Determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics;

[0013] Determining a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication;

[0014] According to the comprehensive optimization communication implementation method, the many-to-many communication operation is completed.

[0015] According to an embodiment of the present application, a network topology feature of a high-performance computer running the application is obtained, and the network topology feature includes:

[0016] The two communicating processes adopt at least one of the multi-level hybrid interconnection modes based on clusters, wherein the multi-level hybrid interconnection modes include intra-node interconnection, primary switching interconnection, secondary switching interconnection and multi-level switching interconnection.

[0017] According to an embodiment of the present application, determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics includes:

[0018] According to the network topology characteristics, the interconnection mode of each point-to-point communication is determined, and according to the interconnection mode of each point-to-point communication, the communication bandwidth capacity of each point-to-point communication is determined; wherein the communication bandwidth capacity of each interconnection mode is a set value or a measured value.

[0019] According to an embodiment of the present application, determining a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication includes:

[0020] Determine the interconnection mode of each point-to-point communication according to the network topology structure characteristics, and determine the communication delay value of each point-to-point communication according to the interconnection mode of each point-to-point communication;

[0021] Determining whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value and the corresponding communication data volume of each point-to-point communication;

[0022] If the point-to-point communication is bandwidth-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a direct sending method. If the point-to-point communication is delay-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a redistribution method between processes of the same level.

[0023] According to an embodiment of the present application, determining whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value, and the corresponding communication data volume of each point-to-point communication includes:

[0024] Dividing the communication data volume by the communication bandwidth capacity to obtain bandwidth communication overhead;

[0025] The ratio of the communication delay value to the bandwidth communication overhead is compared with a preset first ratio threshold. If the ratio is greater than the first ratio threshold, it is determined that the point-to-point communication is delay-limited, otherwise, it is bandwidth-limited.

[0026] According to an embodiment of the present application, in the process of direct sending and redistribution between processes of the same level, the method for determining the order of a process when completing multiple point-to-point communications includes: a random determination method, an interconnection level batch processing method, or an interconnection level round-robin method.

[0027] According to an embodiment of the present application, the method further includes:

[0028] After determining the comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication, the comprehensive optimized communication implementation mode of the many-to-many communication operation of the application is recorded and saved.

[0029] The present application also provides a comprehensive optimization device for a many-to-many communication mode, the device comprising:

[0030] A network feature acquisition module, for acquiring network topology features of a high-performance computer running the application in response to a many-to-many communication operation initiated by a process group of the application;

[0031] A bandwidth capacity determination module, used to determine the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics;

[0032] An implementation mode determination module, used to determine a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication;

[0033] The many-to-many communication module is used to complete the many-to-many communication operation according to the comprehensive optimization communication implementation method.

[0034] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned comprehensive optimization method of the many-to-many communication mode when executing the computer program.

[0035] The present application also provides a computer-readable storage medium, which stores a computer program for executing the above-mentioned comprehensive optimization method of the many-to-many communication mode.

[0036] Compared with the prior art, the beneficial effects that can be achieved by at least one of the above-mentioned technical solutions adopted in the embodiments of this specification include at least the following: the embodiments of the present invention, for the many-to-many collective communication mode, first obtain the network topology characteristics of the high-performance computer running the application; then determine the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation based on the network topology characteristics; finally, determine the comprehensive optimized communication implementation method of the many-to-many communication operation based on the communication bandwidth capacity and the corresponding communication data volume, complete the many-to-many communication operation, and effectively improve the efficiency of the many-to-many communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 It is a flowchart of a comprehensive optimization method for a many-to-many communication mode according to an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of a comprehensive optimization device for a many-to-many communication mode according to an embodiment of the present invention;

[0040] Figure 3 It is a structural schematic diagram of the computer device of the present invention. DETAILED DESCRIPTION

[0041] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0042] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0043] like Figure 1 As shown, an embodiment of the present invention provides a comprehensive optimization method for a many-to-many communication mode, the method comprising:

[0044] S101. In response to a many-to-many communication operation initiated by a process group of an application, obtaining network topology characteristics of a high-performance computer running the application;

[0045] S102. Determine the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology characteristics;

[0046] S103. Determine the comprehensive optimization communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication;

[0047] S104. Complete the many-to-many communication operation according to the comprehensive optimization communication implementation method.

[0048] In one embodiment, a network topology feature of a high-performance computer running the application is obtained, and the network topology feature includes:

[0049] The two communicating processes adopt at least one of the multi-level hybrid interconnection modes based on clusters, wherein the multi-level hybrid interconnection modes include intra-node interconnection, primary switching interconnection, secondary switching interconnection and multi-level switching interconnection.

[0050] Modern high-performance computers often use a cluster-based multi-level hybrid interconnection method to enable massive processor cores to work together. In specific implementation, the multi-level hybrid interconnection method specifically includes:

[0051] (1) Intra-node interconnection, that is, the multiple processor cores in the same computing node are interconnected through shared memory or buses, so that the communication between processes running in the same computing node does not need to go through high-speed switches, thus achieving very fast communication speeds.

[0052] (2) Level 1 switching interconnection, that is, two computing nodes can communicate with each other through only one switch. In many high-performance computer configurations, level 1 switching interconnection of computing nodes in a cabinet can usually be achieved, where each computing node can have one or even two network cables connected to the level 1 switch. Since communications under level 1 switching interconnection need to pass through high-speed switches, the communication speed will be limited by the capabilities of the high-speed switches.

[0053] (3) Level 2 switching interconnection, that is, two computing nodes need to go through two layers (usually three) of switches to complete communication with each other. Large high-performance computers need to interconnect multiple cabinets. At this time, at least level 2 switching interconnection is required, in which several level 1 switches are connected to one level 2 switch. Although a level 1 switch can have more than one network cable connected to the level 2 switch, the number of network cables will be significantly less than the number of computing nodes connected to the level 1 switch. Therefore, compared with level 1 switching interconnection, the communication delay between two computing nodes under level 2 switching interconnection will increase exponentially and the average communication bandwidth capacity will decrease exponentially.

[0054] (4) Multi-level switching interconnection. This allows the number of computing nodes on high-performance computers to increase exponentially based on the form of a binary tree. Based on the two-level switching interconnection, the communication delay between two computing nodes under multi-level interconnection will further increase exponentially and the average communication bandwidth capacity will further decrease exponentially.

[0055] On high-performance computers, each computing node has a unique node name (or number), and the application can obtain the computing node name (or number) running each process through the corresponding program interface. Therefore, for two processes, by comparing the corresponding computing node names (or numbers), you can know the interconnection between the two processes. For example, when the computing node names (or numbers) of two processes are the same, the two processes belong to intra-node interconnection; when the computing node names (or numbers) of two processes are different but the high fields are the same, the two processes belong to primary switching interconnection.

[0056] In one embodiment, determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics includes:

[0057] According to the network topology characteristics, the interconnection mode of each point-to-point communication is determined, and according to the interconnection mode of each point-to-point communication, the communication bandwidth capacity of each point-to-point communication is determined; wherein the communication bandwidth capacity of each interconnection mode is a set value or a measured value.

[0058] In this embodiment, a many-to-many communication operation is essentially composed of several point-to-point communications, wherein one point-to-point communication has a sending process and a receiving process. Therefore, after obtaining the network topology structure characteristics, the interconnection mode of a point-to-point communication can be obtained, thereby determining the communication bandwidth capacity of the point-to-point communication. In this embodiment, the communication bandwidth capacity value under each interconnection mode can be recorded; the communication bandwidth capacity value of each interconnection mode can be set by an administrator, or can be given by actual measurement of a benchmark test case. Generally speaking, the average communication bandwidth capacity value of the interconnection within a node may reach tens of GB / s, the average communication bandwidth capacity value of the first-level switching interconnection may reach several GB / s, the average communication bandwidth capacity value of the second-level switching interconnection may be around 1GB / s, and the average communication bandwidth capacity of the multi-level switching interconnection is even lower.

[0059] In specific implementation, while determining the communication bandwidth capacity, the communication delay can also be determined according to the interconnection mode of each point-to-point communication. In this embodiment, the communication delay value under each interconnection mode can be recorded; the communication delay value of each interconnection mode can be set by an administrator, or can be given by actual measurement of a benchmark test case. Generally speaking, the communication delay value of the interconnection within the node can be as small as ten nanoseconds, the communication delay value of the first-level switching interconnection may reach several microseconds, and the communication delay value of the second-level switching interconnection and the multi-level switching interconnection is even greater.

[0060] In one embodiment, determining the comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication includes:

[0061] Determine the interconnection mode of each point-to-point communication according to the network topology structure characteristics, and determine the communication delay value of each point-to-point communication according to the interconnection mode of each point-to-point communication;

[0062] Determining whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value and the corresponding communication data volume of each point-to-point communication;

[0063] If the point-to-point communication is bandwidth-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a direct sending method. If the point-to-point communication is delay-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a redistribution method between processes of the same level.

[0064] Among them, according to the communication bandwidth capacity, the communication delay value and the corresponding communication data volume of each point-to-point communication, determining whether the point-to-point communication is bandwidth-limited or delay-limited includes: dividing the communication data volume by the communication bandwidth capacity to obtain the bandwidth communication overhead; comparing the ratio of the communication delay value to the bandwidth communication overhead with a preset first ratio threshold, if the ratio is greater than the first ratio threshold, determining that the point-to-point communication is delay-limited, otherwise, it is bandwidth-limited.

[0065] When this embodiment is implemented, first, it is determined whether each point-to-point communication is bandwidth-limited or delay-limited. For a point-to-point communication, it can be determined whether the point-to-point communication is bandwidth-limited or delay-limited based on its communication bandwidth capacity, communication delay, and communication data volume. In specific implementation, the bandwidth communication overhead can be obtained by dividing the communication data volume by the communication bandwidth capacity, and then the communication delay and the bandwidth communication overhead can be compared to determine whether it is bandwidth-limited or delay-limited. For example, when the ratio of the communication delay to the bandwidth communication overhead is greater than a preset first ratio threshold, it is determined to be delay-limited. The first ratio threshold under different interconnection modes may be different, that is, the first ratio threshold of the first-level switching interconnection, the second-level switching interconnection, and the multi-level switching interconnection may be different. The point-to-point communications interconnected within the node can all be considered to be bandwidth-limited.

[0066] Then, the processing mode of each pending point-to-point communication is determined, and the processing mode includes direct sending and sending after redistribution between processes at the same level. Among them, when a pending point-to-point communication is bandwidth-limited, its processing mode is direct sending. The pending point-to-point communication other than the direct sending mode will be processed in a redistribution mode between processes at the same level. The redistribution sending mode between processes at the same level redistributes the communication data between several delay-limited pending point-to-point communications, and then obtains several new pending point-to-point communications, with the purpose of making each new pending point-to-point communication bandwidth-limited. The redistribution sending mode between processes at the same level of the first-level switching interconnection will be implemented based on the data communication between processes in the same node; the redistribution sending mode between processes at the same level of the second-level switching interconnection will be implemented based on the data communication between processes in the same node, and based on the data communication between processes in the same first-level switch; the redistribution sending mode between processes at the same level of the multi-level switching interconnection is similar.

[0067] Specifically, for several pending point-to-point communications of the first-level switching interconnection, they will first be divided into several small groups, where all the sending processes of the point-to-point communications in each group are in the same computing node. The sending processes of all the point-to-point communications in the same group exchange data with each other through communication, and obtain several new pending point-to-point communications, where the combination formed by the sending node number and the receiving node number of each new pending point-to-point communication is usually unique among all new pending point-to-point communications. Redistribution between processes at the same level can enable multiple point-to-point communications between two computing nodes to be merged into a few or even one point-to-point communication, thereby having the opportunity to make delay-limited communications become bandwidth-limited.

[0068] For several pending point-to-point communications of the secondary exchange interconnection, they can be first redistributed among processes in each corresponding computing node, and then redistributed among processes in a group of computing nodes corresponding to each corresponding primary switch as needed, so that multiple point-to-point communications between two computing nodes can be merged into a few or even one point-to-point communication, and multiple point-to-point communications between two primary switches can be merged into a few or even one point-to-point communication. When a new pending point-to-point communication formed by the redistribution among processes in a computing node is bandwidth-limited, there is no need to redistribute the point-to-point communication among processes.

[0069] A new point-to-point communication formed based on the redistribution between processes may contain communication data to be received by multiple processes. After receiving such communication data, a process needs to send each part of the data to the corresponding computing node or process.

[0070] In one embodiment, during the direct sending and the redistribution between peer processes, the order of a process completing multiple point-to-point communications is determined in a manner including: random determination, interconnection level batch processing, or interconnection level round-robin.

[0071] Both direct sending and redistribution between peer processes will introduce multiple point-to-point communications, which may belong to different interconnection methods. In specific implementation, when a process completes multiple point-to-point communications, there will always be a sequence, which can be determined in the following ways:

[0072] (1) Random method: a point-to-point communication is randomly selected and sent each time.

[0073] (2) Interconnection level batch processing mode. That is, after completing the sending of all point-to-point communications of one interconnection level, the point-to-point communications of another interconnection level are sent.

[0074] (3) Interconnection level rotation method. There are multiple rounds, in which after completing the sending of all point-to-point communications of one interconnection level in each round, the point-to-point communications of another interconnection level are sent. The number of interconnection levels involved and the predicted total time overhead of point-to-point communications of each level should be as close as possible between multiple rounds.

[0075] Generally speaking, the interconnection level rotation method has a greater chance of overlapping point-to-point communications at different interconnection levels, which helps to fully utilize the parallelism of the communication grid and reduce the competition for communication hardware resources between the first-level switching interconnection, the second-level switching interconnection and the multi-level switching interconnection point-to-point communications.

[0076] In one embodiment, the method further includes: after determining the comprehensive optimized communication implementation method of the many-to-many communication operation based on the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication, recording and saving the comprehensive optimized communication implementation method of the many-to-many communication operation of the application.

[0077] In this embodiment, since an application usually calls the same many-to-many communication operation multiple times during one operation, although determining the comprehensive optimized communication implementation of a many-to-many communication operation will introduce additional overhead, when the same many-to-many communication operation is called again, the previously determined comprehensive optimized communication implementation of the many-to-many communication operation can be used; that is, in the process of determining the comprehensive optimized communication implementation of a many-to-many communication operation, the comprehensive optimized communication is recorded.

[0078] like Figure 2 As shown, the present application also provides a comprehensive optimization device 200 for a many-to-many communication mode, the device comprising:

[0079] A network feature acquisition module 201, for acquiring network topology features of a high-performance computer running the application in response to a many-to-many communication operation initiated by a process group of the application;

[0080] The bandwidth capacity determination module 202 is used to determine the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics;

[0081] An implementation mode determination module 203 is used to determine a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication;

[0082] The many-to-many communication module 204 is used to complete the many-to-many communication operation according to the comprehensive optimization communication implementation method.

[0083] When the device is implemented specifically, the network topology structure features include: two communicating processes adopt at least one of the multi-level hybrid interconnection methods based on clusters, and the multi-level hybrid interconnection methods include intra-node interconnection, first-level switching interconnection, second-level switching interconnection and multi-level switching interconnection.

[0084] In specific implementation, the bandwidth capacity determination module 202 is used to determine the interconnection mode of each point-to-point communication according to the network topology structure characteristics, and determine the communication bandwidth capacity of each point-to-point communication according to the interconnection mode of each point-to-point communication; wherein the communication bandwidth capacity of each interconnection mode is a set value or a measured value.

[0085] In specific implementation, the implementation mode determination module 203 is used to determine the interconnection mode of each point-to-point communication according to the network topology structure characteristics, and determine the communication delay value of each point-to-point communication according to the interconnection mode of each point-to-point communication; determine whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value and the corresponding communication data volume of each point-to-point communication; if the point-to-point communication is bandwidth-limited, determine that the comprehensive optimized communication implementation mode of the point-to-point communication is a direct sending mode; if the point-to-point communication is delay-limited, determine that the comprehensive optimized communication implementation mode of the point-to-point communication is a redistribution mode between processes of the same level.

[0086] Among them, the communication data volume is divided by the communication bandwidth capacity to obtain the bandwidth communication overhead; the ratio of the communication delay value to the bandwidth communication overhead is compared with a preset first ratio threshold, and if the ratio is greater than the first ratio threshold, it is determined that the point-to-point communication is delay-limited, otherwise, it is bandwidth-limited.

[0087] Among them, in the process of direct sending and redistribution between processes of the same level, the order of a process in completing multiple point-to-point communications includes: random determination, interconnection level batch processing, or interconnection level round-robin.

[0088] In specific implementation, the many-to-many communication module 204 is further used to record and save the comprehensive optimized communication implementation mode of the many-to-many communication operation of the application after determining the comprehensive optimized communication implementation mode of the many-to-many communication operation.

[0089] In one embodiment, a computer device is provided, such as Figure 3 As shown, it includes a memory 301, a processor 302, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned comprehensive optimization method of the many-to-many communication mode is implemented.

[0090] Specifically, the computer device may be a computer terminal, a server or a similar computing device.

[0091] In this embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned comprehensive optimization method of the many-to-many communication mode.

[0092] Specifically, computer-readable storage media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable storage media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0093] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of the present invention are not limited to any specific combination of hardware and software.

[0094] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A comprehensive optimization method for many-to-many communication mode, characterized in that: The method comprises: In response to a many-to-many communication operation initiated by a process group of an application, obtaining network topology characteristics of a high-performance computer running the application; Determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology characteristics; Determining a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication; According to the comprehensive optimization communication implementation method, the many-to-many communication operation is completed.

2. The comprehensive optimization method of many-to-many communication mode according to claim 1, characterized in that: Acquire network topology characteristics of a high-performance computer running the application, wherein the network topology characteristics include: The two communicating processes adopt at least one of the multi-level hybrid interconnection modes based on clusters, wherein the multi-level hybrid interconnection modes include intra-node interconnection, primary switching interconnection, secondary switching interconnection and multi-level switching interconnection.

3. The comprehensive optimization method of many-to-many communication mode according to claim 1, characterized in that: Determining the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics includes: According to the network topology characteristics, the interconnection mode of each point-to-point communication is determined, and according to the interconnection mode of each point-to-point communication, the communication bandwidth capacity of each point-to-point communication is determined; wherein the communication bandwidth capacity of each interconnection mode is a set value or a measured value.

4. The comprehensive optimization method of many-to-many communication mode according to claim 1, characterized in that: Determining a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication, including: Determine the interconnection mode of each point-to-point communication according to the network topology structure characteristics, and determine the communication delay value of each point-to-point communication according to the interconnection mode of each point-to-point communication; Determining whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value and the corresponding communication data volume of each point-to-point communication; If the point-to-point communication is bandwidth-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a direct sending method. If the point-to-point communication is delay-limited, the comprehensive optimized communication implementation method of the point-to-point communication is determined to be a redistribution method between processes of the same level.

5. The comprehensive optimization method of many-to-many communication mode according to claim 4 is characterized in that: Determining whether the point-to-point communication is bandwidth-limited or delay-limited according to the communication bandwidth capacity, the communication delay value, and the corresponding communication data volume of each point-to-point communication includes: Dividing the communication data volume by the communication bandwidth capacity to obtain bandwidth communication overhead; The ratio of the communication delay value to the bandwidth communication overhead is compared with a preset first ratio threshold. If the ratio is greater than the first ratio threshold, it is determined that the point-to-point communication is delay-limited, otherwise, it is bandwidth-limited.

6. The comprehensive optimization method of many-to-many communication mode according to claim 4, characterized in that: In the process of direct sending and redistribution between processes at the same level, the order of a process completing multiple point-to-point communications includes: random determination, interconnection level batch processing, or interconnection level round-robin.

7. The comprehensive optimization method of many-to-many communication mode according to claim 1, characterized in that: The method further comprises: After determining the comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication, the comprehensive optimized communication implementation mode of the many-to-many communication operation of the application is recorded and saved.

8. A comprehensive optimization device for many-to-many communication mode, characterized in that: The device comprises: A network feature acquisition module, for acquiring network topology features of a high-performance computer running the application in response to a many-to-many communication operation initiated by a process group of the application; A bandwidth capacity determination module, used to determine the communication bandwidth capacity of each point-to-point communication in the many-to-many communication operation according to the network topology structure characteristics; An implementation mode determination module, used to determine a comprehensive optimized communication implementation mode of the many-to-many communication operation according to the communication bandwidth capacity and the corresponding communication data volume of each point-to-point communication; The many-to-many communication module is used to complete the many-to-many communication operation according to the comprehensive optimization communication implementation method.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the comprehensive optimization method of the many-to-many communication mode described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the comprehensive optimization method of the many-to-many communication mode according to any one of claims 1 to 7.