CN120560921B - Communication performance test method, device and storage medium - Google Patents

Abstract

The application discloses a communication performance testing method, equipment and a storage medium, which relate to the technical field of communication and comprise the steps of constructing a testing link comprising a second type processor of a server, a network port of the server, a second type processor of a client and a network port of the client according to the topological relation of peripheral component interconnect standard equipment corresponding to a host after network configuration is completed, and executing remote direct memory access test of a single-machine graphic processor on the testing link. By adopting the technical scheme, the test can be completed on a single machine, the server and the client are not required to be arranged on different computing nodes, the computing resources are saved, meanwhile, the test links are not required to be manually configured, the unified configuration of the processor model, the network card drive and the like of different computing nodes is also not required, and the configuration requirement is reduced.

Description

Communication performance test method, device and storage medium

Technical Field

The present application relates to the field of communications technologies, and in particular, to a method, an apparatus, and a storage medium for testing communications performance.

Background

Graphics processors (Graphics Processing Unit, GPUs) are specially designed for graphics rendering, have a large number of computing cores, and are suitable for processing deep learning and other high-performance computing tasks, and the communication overhead in training of a deep learning model and other high-performance computing scenes is relatively large, so that a large amount of data interaction is required between GPUs of different computing nodes. The GPU remote direct memory access (GPU DIRECT RDMA, GDR) technology can enable the GPU using the high-speed serial computer expansion bus standard (PERIPHERAL COMPONENT INTERCONNECT EXPRESS, PCIe) to directly exchange data with the third-party equipment, and a central processing unit (Central Processing Unit, CPU) is not needed, so that the increase of communication delay caused by the bottleneck of CPU performance is avoided, and the communication performance is obviously improved.

Before the GDR is formally started, in order to adapt to the complex environment dependence in actual deployment, the potential performance traps are avoided, and the performance test is required to be performed first. The current GDR performance test method is mainly distributed remote direct memory access (Remote Direct Memory Access, RDMA) bandwidth benchmark test, and needs to set a server and a client on different computing nodes and perform multi-machine GDR test. However, the requirement of multi-machine GDR test on hardware compatibility is high, multiple-node GPU models, network card drivers and the like are required to be unified, and more test resources are required, so that the waste and redundancy of the test resources are easy to cause. Therefore, how to reduce the requirements of the GDR test on hardware resources is a technical problem to be solved currently.

Disclosure of Invention

The application provides a communication performance testing method, equipment and a storage medium, which at least solve the problem that the GDR test in the related technology has higher requirement on hardware resources.

The application provides a communication performance test method, which comprises the following steps:

The method comprises the steps that a topological relation of peripheral component interconnect standard equipment corresponding to a host is obtained, the host comprises at least two first type processors, and the peripheral component interconnect standard equipment comprises a second type processor and a network port;

network configuration is carried out on the network ports in the host machine based on a preset configuration rule;

The method comprises the steps that a test link is constructed based on a topological relation, the test link comprises a server-side second-type processor, a server-side network port, a client-side second-type processor and a client-side network port, the server-side second-type processor and the server-side network port are slave devices of a first processor in at least two first-type processors, and the client-side second-type processor and the client-side network port are slave devices of a second processor in at least two first-type processors;

based on the test link, remote direct memory access test of the graphic processor is executed in the host machine, and a test result is obtained.

The application also provides a communication performance testing device, which comprises:

the first acquisition module is used for acquiring the topological relation of the peripheral component interconnection standard equipment corresponding to the host machine, the host machine comprises at least two first type processors, and the peripheral component interconnection standard equipment comprises a second type processor and a network port;

The configuration module is used for carrying out network configuration on the network port in the host machine based on a preset configuration rule;

The first construction module is used for constructing a test link based on a topological relation, wherein the test link comprises a server-side second-type processor, a server-side network port, a client-side second-type processor and a client-side network port, the server-side second-type processor and the server-side network port are slave devices of a first processor in at least two first-type processors, and the client-side second-type processor and the client-side network port are slave devices of a second processor in at least two first-type processors;

and the first test module is used for executing remote direct memory access test of the graphic processor in the host based on the test link to obtain a test result.

The application also provides electronic equipment which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for realizing any step of the communication performance testing method when executing the computer program.

The present application also provides a computer readable storage medium having a computer program stored therein, wherein the computer program when executed by a processor implements the steps of any of the communication performance testing methods described above.

The application also provides a computer program product comprising a computer program which when executed by a processor implements the steps of any of the above communication performance testing methods.

After the network configuration is completed, a test link comprising a second type processor of a server side, a network port of the server side, a second type processor of a client side and the network port of the client side is constructed according to the topological relation of the peripheral component interconnection standard equipment corresponding to the host machine, and a remote direct memory access test of the single-machine graphic processor is executed on the test link. By adopting the technical scheme, the test can be completed on a single machine, the server and the client are not required to be arranged on different computing nodes, the computing resources are saved, meanwhile, the test links are not required to be manually configured, the unified configuration of the processor model, the network card drive and the like of different computing nodes is also not required, and the configuration requirement is reduced.

Drawings

For a clearer description of embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described, it being apparent that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to the drawings without inventive effort for those skilled in the art.

Fig. 1 is a flow chart of a communication performance testing method according to an embodiment of the present application;

Fig. 2 is a flow chart of a method for constructing a test link according to an embodiment of the present application;

FIG. 3 is a flowchart of a method for determining a portal corresponding to a second type processor according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a communication performance testing apparatus according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without making any inventive effort are within the scope of the present application.

It should be noted that in the description of the present application, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. The terms "first," "second," and the like in this specification are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.

The present application will be further described in detail below with reference to the drawings and detailed description for the purpose of enabling those skilled in the art to better understand the aspects of the present application.

Fig. 1 is a flowchart of a communication performance testing method according to an embodiment of the present application, where the method may be performed by an electronic device, and the electronic device may be exemplarily understood as a device such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, a smart tv, etc. As shown in fig. 1, the method provided in this embodiment includes the following steps:

s101, obtaining a topological relation of peripheral component interconnect standard equipment corresponding to a host machine, wherein the host machine comprises at least two first type processors, and the peripheral component interconnect standard equipment comprises a second type processor and a network port.

In one example, before obtaining the topology relationship of the peripheral component interconnect standard device corresponding to the host, the method includes:

The method comprises the steps of deploying a remote direct memory access tool kit of a graphics processor in an operating system of a host machine, deploying a first component corresponding to a second type processor, verifying whether deployment is successful, deploying a second component corresponding to a network card, and verifying whether deployment is successful.

Specifically, the remote direct memory access toolkit (GDR-Toolkits) of the graphics processor can be deployed directly on a host machine, or can be deployed in a remote control manner, the toolkit is copied to an operating system of the host machine, after the deployment is successful, the configuration is carried out under a deployment toolkit catalog, a corresponding first component is deployed for a second type processor, including installing a driving and computing unified device architecture (Compute Unified Device Architecture, CUDA) corresponding to the second type processor, verifying whether the deployment is successful or not through corresponding commands, for example, the verification commands can be nvidia-smai and nvcc-V, after the verification is passed, a nvidia-peer-mem module is deployed and a nvidia-peer-mem module can be successfully loaded, and because a network port is arranged on the network card, the network port function normally operates, the deployment and configuration are required for the network card, the corresponding second component is deployed for the network card device, including installing OpenFabrics Enterprise Distribution (OFED) driving and Open Fabrics Enterprise Distribution Performance Tests (perftest) toolset, and verifying whether the deployment is successful or not. By adopting the technical scheme, the components required by the test can be deployed efficiently and rapidly, the original manually-executed deployment process is converted into automatic calling and execution, and the labor is saved, so that the important task of complex test is overcome.

In this embodiment, the host is configured to execute the GDR test, where the host includes at least two first type processors, peripheral component interconnect standard (PERIPHERAL COMPONENT INTERCONNECT, PCI) devices are set and managed under the first type processors, the first type processors may be CPUs, the second type processors may be GPUs, the network card is an RDMA network card, and the GPUs on the host and the RDMA network card need to be connected through a high bandwidth PCIe bus.

The topology relationship of the PCI device corresponding to the host may be obtained from a configuration file of the PCI device obtained in advance, or may be obtained by executing a command (for example lspci) for obtaining the topology relationship of the PCI device, or may be obtained by a file storing information of the PCI device under a specific directory in the file system, which is not limited herein.

In one example, obtaining a topology relationship of a peripheral component interconnect standard device corresponding to a host includes:

the method comprises the steps of obtaining device addresses of a plurality of peripheral component interconnection standard devices contained in a host and structural relations among the peripheral component interconnection standard devices, and constructing topological relations of the peripheral component interconnection standard devices based on the device addresses and the structural relations.

Specifically, the device address may be a PCI bus-device-function (PCI BDF) address, the structural relationship may be a PCI device tree structure relationship, including a topology structure of a bus, a device, and a function, and after the device address and the structural relationship are obtained, printing is performed according to a topology hierarchy, and the PCI device may further include a network card, a physical high-speed serial computer expansion bus standard switch (PHYSICAL PCIE SWITCH, PHYSW), and a logical high-speed serial computer expansion bus standard switch (Logic PCIE SWITCH, LOGICSW) which are listed step by step respectively in the hierarchies of PhySW, logicSW, a second type processor device, a network card device, and a network port device, so as to construct the topology relationship of the PCI device.

S102, carrying out network configuration on the network port in the host based on a preset configuration rule.

Specifically, network configuration is performed on a network card, the network card is accessed to a switch network, a network card driver is automatically adapted according to the network card model, RDMA is installed, the correct version of the RDMA driver is checked, corresponding configuration rules are selected from preset configuration rules corresponding to various types according to the type of the switch connected with the network port, network port parameters of a host are configured based on the configuration rules through a command line tool, for example, an IP address is configured for each network port in a self-increasing mode according to a preset static Internet protocol (Internet Protocol, IP) address prefix static_ip_prefix, a subnet mask is configured uniformly, and the network port is enabled after configuration is completed.

In one example, network configuration is performed on a network port in a host based on a preset configuration rule, including:

creating a network naming space based on the number of network ports contained in the host machine, adding the network ports to the network naming space correspondingly, and configuring an Internet protocol address for the network ports.

Before the remote direct memory access test of the graphic processor is executed in the host based on the test link to obtain the test result, the method further comprises the following steps:

And performing connectivity test on any two configured network ports.

Specifically, the network ports and network namespaces (namespace) may be in one-to-one correspondence, a corresponding number of network namespaces are created according to the number of network ports included in the host machine, the naming mode of the network namespaces may be < namespaces > < network card names > < network port names >, for example, ns.rdma0.mlx5_0.ens1, after creation is completed, each network port is added to each created network namespace, so as to isolate network application, so that each network port device has independent routing, iptable policy and device management, a stand-alone GDR test may be performed, after addition, an IP address is configured for each network port in each network namespace, and before formally starting the GDR test, connectivity between each network port in the network namespace and each network port in other network namespaces is tested, for example, ns.rdma0.mlx5_0.ens1 ping ns.rdma2.mlx5_2.ens65 is ensured, so that all network ports can be mutually connected. By adopting the technical scheme, the network ports can be isolated through the network naming space, so that each network port device has independent resources, resource contention among the network ports is avoided, communication delay is reduced, and overall performance is improved.

S103, constructing a test link based on a topological relation, wherein the test link comprises a server-side second-type processor, a server-side network port, a client-side second-type processor and a client-side network port, the server-side second-type processor and the server-side network port are slave devices of a first processor in at least two first-type processors, and the client-side second-type processor and the client-side network port are slave devices of a second processor in at least two first-type processors.

The first processor and the second processor may be any first type processor in the host, and the first processor and the second processor are different first type processors, and one or more of the first processor and the second processor may be provided, which is not limited herein.

Specifically, after the topological relation of the PCI devices is acquired, the PCI devices related to the GDR test are determined based on the topological relation, and a test link is constructed according to the related PCI devices, because the client and the server need to be arranged on different first type processors, and the second type processors and the network ports used by the client and the server respectively need to be arranged on the same first type processor, when the test link is constructed, the second type processor of the server and the network port of the server in the test link are the slave devices of the first processor in at least two first type processors on the host, and the second type processor of the client and the network port of the client in the test link are the slave devices of the second processor in at least two first type processors.

S104, based on the test link, executing remote direct memory access test of the graphic processor in the host machine to obtain a test result.

Specifically, after determining the test links, performing a GDR test on each test link to obtain a test result corresponding to each test link, where the test result may include a bandwidth, a delay, a packet loss rate, and the like, which is not limited herein.

In one example, the test conditions of the remote direct memory access test of the graphics processor include a test direction and a test type, the test direction includes a unidirectional test and a bidirectional test, the test type includes a write bandwidth test, a send bandwidth test and a read bandwidth test, the remote direct memory access test of the graphics processor is performed in the host based on the test link, a test result is obtained, and the method includes:

Aiming at each test link in at least one test link, testing based on each test condition to obtain a corresponding condition test result of each test link under each test condition; and integrating the condition test results corresponding to each test link to obtain a test result.

Specifically, the test direction and the test type may be combined to obtain a plurality of test conditions, and for each test link in at least one test link, a test instruction corresponding to each test condition is executed, where the test instruction may be GDR unidirectional write bandwidth test (unidirectional write bandwidth test), GDR bidirectional write bandwidth test (bidirectional write bandwidth test), GDR unidirectional send bandwidth test (unidirectional transmission bandwidth test), GDR bidirectional send bandwidth test (bidirectional transmission bandwidth test), GDR unidirectional read bandwidth test (unidirectional read bandwidth test), GDR bidirectional read bandwidth test (bidirectional read bandwidth test), for example. After obtaining the condition test result of each test link under each test condition, comparing each test result with a preset reference value, judging whether the condition test result of the test link under the test condition is successful, if the condition test result of the test link under each test condition is successful, determining that the test result of the test link is successful, otherwise, determining that the test result of the test link is failed.

In one example, after obtaining the topology relationship of the peripheral component interconnect standard device corresponding to the host, the method further includes:

Constructing a topological graph based on the topological relation;

the method comprises the steps of integrating the condition test results corresponding to each test link, and after the test results are obtained, the method further comprises the following steps:

and marking the target test link in the topological graph when the condition test result and/or the test result corresponding to the target test link are failed.

Specifically, after the topology relationship is obtained, a topology graph may be further drawn and displayed, where the topology graph may include a PCI BDF address and a structural relationship of a PCI device, and may further include a Bus ID of a PCIe Bus, a Bus width, a Bus rate, an upstream PCIE SWITCH location, a network naming space of a network port, and a network port connection state (i.e., link status UP or Down), and when the PCI device further includes a network card, phySW, and LogicSW, the topology graph may further include a physical network card, a network interface, a physical network port, and a complete link of an IP address network naming space. After the test result is obtained, the test result can be marked on the topological graph, specifically, when the condition test result corresponding to the target test link is failure, the target test link can be marked on the topological graph in a highlighting way, meanwhile, the test condition of the test failure is marked, when the test result corresponding to the target test link is failure, the target test link can be marked on the topological graph in a highlighting way, and when the corresponding condition test result and the test result are failure, the target test link can be marked on the topological graph in a highlighting way, and different marking ways can be adopted under different conditions, so that the minute-level log analysis is compressed into second-level visual perception, and the quick positioning of the abnormal result is completed. Optionally, for the links with test failure, the relevant buses can be positioned according to the test links, and the bus width and the bus rate of the relevant buses are compared with the preset bus width and the preset bus rate, so that the bandwidth bottleneck is accurately positioned, and the primary analysis of the reasons of test failure is realized.

After network configuration is completed, a test link comprising a second type processor of a server, a network port of the server, a second type processor of a client and a network port of the client is constructed according to the topological relation of peripheral component interconnection standard equipment corresponding to a host, and a remote direct memory access test of a single-machine graphic processor is executed on the test link. By adopting the technical scheme, the test can be completed on a single machine, the server and the client are not required to be arranged on different computing nodes, the computing resources are saved, meanwhile, the test links are not required to be manually configured, the unified configuration of the processor model, the network card drive and the like of different computing nodes is also not required, and the configuration requirement is reduced.

Fig. 2 is a flowchart of a method for constructing a test link according to an embodiment of the present application, where a first type of processor is a central processor and a second type of processor is a graphics processor. As shown in fig. 2, the method provided in this embodiment includes the following steps:

s201, obtaining mapping relations between a plurality of peripheral component interconnect standard devices contained in the host machine and at least two central processing units.

Specifically, because Non-uniform memory access (Non-Uniform Memory Access Node, NUMA) nodes in the same host machine are in one-to-one correspondence with the central processor, affinity printing can be performed on a plurality of PCI devices contained in the host machine, and the mapping relation between the graphics processor and the network port device and the NUMA nodes is determined, so that in the subsequent test, the second type processor and the network port of the server side, the second type processor of the client side and the network port of the client side in the test link are ensured to be positioned in the same NUMA domain, and belong to the slave devices of the same first type processor.

S202, determining a first processor and a second processor from at least two central processing units.

Specifically, the central processing unit on which the GPU is mounted may be screened out from at least two CPUs, and a part of the CPUs are determined to be the first processor, and another part of the CPUs are determined to be the second processor, optionally, when determining the first processor and the second processor, the first processor and the second processor may be determined according to the number of GPUs downloaded by the CPUs in a manner that the number of GPUs downloaded by the first processor and the second processor is as balanced as possible, for example, 5 GPUs are mounted on CPU1, 3 GPUs are mounted on CPU2, 3 GPUs are mounted on CPU3, CPU1 may be determined to be the first processor, and CPU2 and CPU3 may be determined to be the second processor.

S203, traversing a first graphic processor subordinate to the first processor and a second graphic processor subordinate to the second processor, and combining each first graphic processor and each second graphic processor to obtain a server-side second type processor and a client-side second type processor in at least one test link.

Specifically, after determining the first processor and the second processor, traversing the first graphics processor under the first processor and the second graphics processor under the second processor, pairing and combining each first graphics processor with each second graphics processor, if all the first GPU/second GPU are configured completely in the first graphics processor and the second graphics processor, and unpaired GPU exists in the second GPU/first GPU, selecting the GPU from the paired first GPU/second GPU to pair with the unpaired second GPU/first GPU, combining the paired first GPU and second GPU as a server second type processor and a client second type processor in one test link, and obtaining the test link corresponding to the group number.

S204, determining a target service end network port corresponding to the target service end second type processor and a target client end network port corresponding to the target client end second type processor according to the target service end second type processor and the target client end second type processor contained in each test link in at least one test link based on the topological relation.

Specifically, after determining the second type processor of the target server and the second type processor of the target client contained in each test link, determining a target server portal matched with the second type processor of the target server and a target client portal matched with the second type processor of the target client further based on the topological relation.

According to the communication performance testing method provided by the embodiment of the application, the mapping relation between a plurality of peripheral component interconnection standard devices contained in a host machine and at least two central processing units is obtained, the first processor and the second processor are determined from the at least two central processing units, the first graphic processor subordinate to the first processor and the second graphic processor subordinate to the second processor are traversed, the first graphic processors and the second graphic processors are combined to obtain the second type processor of the service end and the second type processor of the client in at least one testing link, the second type processor of the target service end and the second type processor of the target client contained in each testing link are determined based on the topological relation, the network port of the target service end corresponding to the second type processor of the target service end and the network port of the target client corresponding to the second type processor are traversed, the GPU subordinate to different CPUs can be traversed, the communication of the multiple testing links are formed by combining flexibly, the communication of the multiple testing links covering the CPU completely, the performance of the GPU is realized, the automatic configuration of the GPU is reduced, the corresponding hardware configuration of the GPU is ensured, the actual testing environment is ensured, and the reliability of the corresponding testing environment is improved.

Fig. 3 is a flowchart of a method for determining a network port corresponding to a second type processor according to an embodiment of the present application, where the peripheral component interconnect standard device further includes a network card, a physical high-speed serial computer expansion bus standard switch, and a logical high-speed serial computer expansion bus standard switch. As shown in fig. 3, the method provided in this embodiment includes the following steps:

S301, judging whether a first network card exists or not based on a topological relation, wherein the first network card is a network card under a target logic high-speed serial computer expansion bus standard switch to which a target service end second type processor belongs.

And S302, if the network interface exists, determining the first network port of the first network card as a target service end network port.

And S303, if not, judging whether a second network card exists, wherein the second network card is a network card under the target physical high-speed serial computer expansion bus standard switch under the target logical high-speed serial computer expansion bus standard switch.

And S304, if the network port exists, determining a second network port of the second network card as a target service end network port.

And S305, if the network interface does not exist, determining a third network port of a third network card as a target server port, wherein the third network card is a network card subordinate to a root complex to which the target physical high-speed serial computer expansion bus standard switch belongs.

Specifically, when determining the target server port corresponding to the target server second type processor, the first network card under the same target LogicSW with the target server second type processor may be preferentially searched according to the topology relationship, if the first network card exists, the first network port on the first network card is determined to be the target server port, if the first network card does not exist, the second network card under the same target PhySW with the target server second type processor but under different LogicSW is searched in the topology relationship along the target LogicSW, if the second network card exists, the second network port on the second network card is determined to be the target server port, if the second network card does not exist, the third network card under the same Root Complex (Root Complex) with the target server second type processor is searched in the topology relationship along the target PhySW, and the third network port on the third network card is determined to be the target server port, thereby determining the target server port corresponding to the target server second type processor.

Optionally, a specific method for determining the target client portal corresponding to the second type processor of the target client based on the topology relationship is similar to S301-S305, and will not be described herein.

According to the communication performance testing method provided by the embodiment of the application, based on the topological relation, whether the first network card exists is judged, the first network card is the network card of the target logic high-speed serial computer expansion bus standard switch of the target service end, if the first network card exists, the first network port of the first network card is determined to be the network port of the target service end, if the first network card does not exist, whether the second network card exists is judged, the second network card is the network card of the target physical high-speed serial computer expansion bus standard switch of the target logic high-speed serial computer expansion bus standard switch, if the second network card exists, the second network port of the second network card is determined to be the network port of the target service end, if the second network card does not exist, the third network port of the third network card is determined to be the network port of the target service end, and the third network card is the root complex of the target physical high-speed serial computer expansion bus standard switch, the network card of the logic PCIE SWITCH, physical PCIE SWITCH and the hierarchical search strategy of the CPU sequence can be adopted, the complexity of the hardware environment is enhanced, the network port closest to the network port is preferentially selected, the network delay efficiency is improved, the network utilization rate is reduced, and the GPU is compatible, and the network utilization bandwidth is reduced.

From the description of the above embodiments, it will be clear to a person skilled in the art that the method according to the above embodiments may be implemented by means of software plus the necessary general hardware platform, but of course also by means of hardware, but in many cases the former is a preferred embodiment.

The embodiment of the application also provides a communication performance testing device which can be understood as the electronic equipment or part of functional modules in the electronic equipment. As shown in fig. 4, the communication performance test apparatus 400 includes:

A first obtaining module 410, configured to obtain a topology relationship of a peripheral component interconnect standard device corresponding to a host, where the host includes at least two processors of a first type, and the peripheral component interconnect standard device includes a processor of a second type and a network port;

the configuration module 420 is configured to perform network configuration on the network port in the host based on a preset configuration rule;

the first construction module 430 is configured to construct a test link based on a topological relation, where the test link includes a server second type processor, a server network port, a client second type processor, and a client network port, the server second type processor and the server network port are slave devices of a first processor of the at least two first type processors, and the client second type processor and the client network port are slave devices of a second processor of the at least two first type processors;

the first test module 440 is configured to perform remote direct memory access test of the graphics processor in the host based on the test link, to obtain a test result.

In one example, the communication performance test apparatus 400 further includes:

A first deployment module for deploying a graphics processor remote direct memory access toolkit in an operating system of a host;

The second deployment module is used for deploying the first component corresponding to the second type processor and verifying whether the deployment is successful or not;

The third deployment module is used for deploying the second component corresponding to the network card and verifying whether the deployment is successful.

In one example, the first acquisition module 410 includes:

The device comprises an acquisition unit, a storage unit and a control unit, wherein the acquisition unit is used for acquiring device addresses of a plurality of peripheral component interconnection standard devices contained in a host machine and structural relations among the plurality of peripheral component interconnection standard devices;

The construction unit is used for constructing the topological relation of the peripheral component interconnection standard equipment based on the equipment address and the structural relation.

In one example, the first type of processor is a central processor and the second type of processor is a graphics processor, the communication performance test apparatus 400 further includes:

The second acquisition module is used for acquiring the mapping relation between a plurality of peripheral component interconnect standard devices contained in the host machine and at least two central processing units;

The first acquisition module 410 includes:

a first determining unit configured to determine a first processor and a second processor from at least two central processors;

The traversal unit is used for traversing a first graphic processor subordinate to the first processor and a second graphic processor subordinate to the second processor, and combining each first graphic processor and each second graphic processor to obtain a server-side second type processor and a client-side second type processor in at least one test link;

The second determining unit is configured to determine, for a target server second type processor and a target client second type processor included in each test link in at least one test link, a target server portal corresponding to the target server second type processor, and a target client portal corresponding to the target client second type processor based on a topological relation.

In one example, the peripheral component interconnect standard device further includes a network card, a physical high-speed serial computer expansion bus standard switch, and a logical high-speed serial computer expansion bus standard switch, and the second determining unit includes:

the first judging subunit is used for judging whether a first network card exists or not based on the topological relation, wherein the first network card is a network card of a target logic high-speed serial computer expansion bus standard switch to which a target service end second type processor belongs;

The first determining subunit is configured to determine, if the first determining subunit exists, a first network port of the first network card as a target server network port;

The second judging subunit is configured to judge whether a second network card exists if the second network card does not exist, where the second network card is a network card under the target physical high-speed serial computer expansion bus standard switch to which the target logical high-speed serial computer expansion bus standard switch belongs;

the second determining subunit is configured to determine, if the second determining subunit exists, a second network port of the second network card as a target server network port;

And the third determining subunit is configured to determine, if the third determining subunit does not exist, a third network port of the third network card as a target server port, where the third network card is a network card subordinate to the root complex to which the target physical high-speed serial computer expansion bus standard switch belongs.

In one example, the configuration module 420 includes:

the creation unit is used for creating a network naming space based on the number of network ports contained in the host machine;

The adding unit is used for correspondingly adding the network port to the network naming space;

The configuration unit is used for configuring an Internet protocol address for the network port;

The communication performance test apparatus 400 further includes:

and the second testing module is used for testing connectivity of any two configured network ports.

In one example, the test conditions of the graphics processor remote direct memory access test include a test direction and a test type, the test direction includes a unidirectional test and a bidirectional test, the test type includes a write bandwidth test, a transmit bandwidth test, and a read bandwidth test, the first test module 440 further includes:

The test unit is used for testing each test link in at least one test link based on each test condition to obtain a corresponding condition test result of each test link under each test condition;

And the integration unit is used for integrating the condition test results corresponding to each test link to obtain test results.

In one example, the communication performance test apparatus 400 further includes:

the second construction module is used for constructing a topological graph based on the topological relation;

the first test module 440 further comprises:

and the labeling unit is used for labeling the target test link in the topological graph when the condition test result and/or the test result corresponding to the target test link are failed.

The description of the features in the embodiment corresponding to the communication performance testing device may refer to the related description of the embodiment corresponding to the communication performance testing method, which is not described in detail herein.

An embodiment of the application also provides an electronic device comprising a memory having stored therein a computer program and a processor arranged to run the computer program to perform the steps of any of the communication performance test method embodiments described above.

Embodiments of the present application also provide a computer readable storage medium having a computer program stored therein, wherein the computer program is configured to perform the steps of any of the communication performance test method embodiments described above when run.

In an exemplary embodiment, the computer readable storage medium may include, but is not limited to, a U disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a removable hard disk, a magnetic disk, or an optical disk, etc. various media in which a computer program may be stored.

Embodiments of the present application also provide a computer program product comprising a computer program which, when executed by a processor, implements the steps of any of the communication performance test method embodiments described above.

Embodiments of the present application also provide another computer program product comprising a non-volatile computer readable storage medium storing a computer program which when executed by a processor implements the steps of any of the communication performance test method embodiments described above.

Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of functionality in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The method, the device and the storage medium for testing the communication performance provided by the application are described in detail. The principles and embodiments of the present application have been described herein with reference to specific examples, the description of which is intended only to facilitate an understanding of the method of the present application and its core ideas. It should be noted that it will be apparent to those skilled in the art that various modifications and adaptations of the application can be made without departing from the principles of the application and these modifications and adaptations are intended to be within the scope of the application as defined in the following claims.

Claims (9)

1. A communication performance testing method, comprising:

The method comprises the steps that the topological relation of peripheral component interconnect standard equipment corresponding to a host machine is obtained, the host machine comprises at least two first type processors, the peripheral component interconnect standard equipment comprises a second type processor and a network port, the first type processor comprises a central processing unit, and the second type processor comprises a graphic processor;

Network configuration is carried out on the network ports in the host machine based on a preset configuration rule;

Constructing a test link based on the topological relation, wherein the test link comprises a second type processor of a server, a server network port, a second type processor of a client and a client network port, the second type processor of the server and the server network port are slave devices of a first processor of the at least two first type processors, and the second type processor of the client and the client network port are slave devices of a second processor of the at least two first type processors;

based on the test link, executing remote direct memory access test of the graphic processor in the host to obtain a test result;

The network configuration of the network port in the host machine based on the preset configuration rule comprises the following steps:

creating a network namespace based on the number of network ports contained by the host;

Correspondingly adding the network port to the network naming space;

Configuring an Internet protocol address for the network port;

the test conditions of the remote direct memory access test of the graphics processor comprise a test direction and a test type, the test direction comprises a unidirectional test and a bidirectional test, the test type comprises a write bandwidth test, a send bandwidth test and a read bandwidth test, and the remote direct memory access test of the graphics processor is executed in the host based on the test link to obtain a test result, and the method comprises the following steps:

aiming at each test link in at least one test link, testing based on each test condition to obtain a condition test result corresponding to each test link under each test condition;

And integrating the conditional test results corresponding to the test links to obtain the test results.

2. The method according to claim 1, wherein before the obtaining the topology relationship of the peripheral component interconnect standard device corresponding to the host, the method further comprises:

deploying a graphics processor remote direct memory access toolkit in an operating system of the host;

Deploying a first component corresponding to the second type processor, and verifying whether deployment is successful;

and deploying a second component corresponding to the network card, and verifying whether deployment is successful.

3. The method according to claim 1, wherein the obtaining the topology relationship of the peripheral component interconnect standard device corresponding to the host includes:

Acquiring device addresses of a plurality of peripheral component interconnect standard devices contained in the host and structural relations among the plurality of peripheral component interconnect standard devices;

And constructing the topological relation of the peripheral component interconnect standard equipment based on the equipment address and the structural relation.

4. The method of claim 1, wherein prior to said constructing a test link based on said topology, said method further comprises:

obtaining mapping relations between a plurality of peripheral component interconnect standard devices and at least two central processing units contained in the host;

The building the test link based on the topological relation comprises the following steps:

Determining the first processor and the second processor from the at least two central processors;

traversing a first graphic processor subordinate to the first processor and a second graphic processor subordinate to the second processor, and combining each first graphic processor and each second graphic processor to obtain a server-side second type processor and a client-side second type processor in at least one test link;

And determining a target server port corresponding to the target server second type processor and a target client port corresponding to the target client second type processor according to the target server second type processor and the target client second type processor contained in each test link in the at least one test link based on the topological relation.

5. The method of claim 4, wherein the peripheral component interconnect standard device further comprises a network card, a physical high-speed serial computer expansion bus standard switch, and a logical high-speed serial computer expansion bus standard switch, wherein the determining the target server portal corresponding to the target server second type processor based on the topology relationship comprises:

judging whether a first network card exists or not based on the topological relation, wherein the first network card is a network card of a target logic high-speed serial computer expansion bus standard switch to which a second type processor of the target server belongs;

if yes, determining a first network port of the first network card as the target server port;

If not, judging whether a second network card exists, wherein the second network card is a network card of the target physical high-speed serial computer expansion bus standard switch to which the target logical high-speed serial computer expansion bus standard switch belongs;

If yes, determining a second network port of the second network card as the target server port;

if the network interface does not exist, a third network port of a third network card is determined to be the target server port, and the third network card is a network card subordinate to a root complex to which the target physical high-speed serial computer expansion bus standard switch belongs.

6. The method of claim 1, wherein the performing a graphics processor remote direct memory access test in the host based on the test link, prior to obtaining a test result, further comprises:

And performing connectivity test on any two configured network ports.

7. The method according to claim 1, wherein after the obtaining the topology relation of the peripheral component interconnect standard device corresponding to the host, the method further comprises:

Constructing a topological graph based on the topological relation;

and integrating the condition test results corresponding to the test links to obtain the test results, wherein the method further comprises the following steps:

And marking the target test link in the topological graph when the condition test result and/or the test result corresponding to the target test link are failed.

8. An electronic device, comprising:

A memory for storing a computer program;

A processor for implementing the steps of the communication performance testing method according to any one of claims 1 to 7 when executing said computer program.

9. A computer readable storage medium, characterized in that a computer program is stored in the computer readable storage medium, wherein the computer program, when being executed by a processor, implements the steps of the communication performance test method according to any one of claims 1 to 7.