Plug-and-play high-performance GPU computing workstation construction method and system

Abstract

The present invention relates to the field of parallel processing technology, and discloses a plug-and-play high-performance GPU computing workstation construction method and system, including: identifying parallel hardware in supporting hardware, using data flow to analyze the parallel performance of the parallel hardware; calculating the hardware budget of the parallel hardware; querying the construction area of ​​the connected workstation, simulating the path connection relationship between the connected workstation and the regional network in the construction area, calculating the path consumption of the path connection relationship, and determining the final connection path between the connected workstation and the regional network in the path connection relationship; using the data type to be processed to generate test data for the connected workstation, and constructing a parallel correlation module for the connected workstation; performing data testing on the connected workstation, the final connection path, and the parallel correlation module, and completing the workstation construction using the connected workstation, the final connection path, and the parallel correlation module. The present invention takes into account the performance after the workstation is connected to the network and the selection of parallel data.

Description

Technical Field

[0001] The present invention relates to a method and system for constructing a plug-and-play high-performance GPU computing workstation, and belongs to the technical field of parallel processing. Background Art

[0002] A high-performance GPU computing workstation is specifically designed for tasks that require extremely high computational loads, such as graphics processing, scientific computing, and machine learning. In addition, a workstation connected to the network is a complete workstation, which can mobilize other resources in the network world, such as data processing resources, data storage resources, etc., to achieve high-performance data processing of the GPU.

[0003] First, the existing technology considers relatively simple things when building a workstation, such as only considering the cost of purchasing hardware and whether the hardware can realize data processing and the performance of data processing. However, a workstation is not an island, which means that a single workstation must be connected to other computer systems to realize distributed processing of tasks. When establishing a connection between the workstation and other computer systems, the connection path problem must be considered. Secondly, when using the workstation for parallel data processing, it is necessary to pre-select which cases to process in parallel. However, the existing technology lacks consideration of the performance of the workstation after connecting to the network and the selection of parallel data. Summary of the Invention

[0004] The present invention provides a method and system for constructing a plug-and-play high-performance GPU computing workstation, the main purpose of which is to consider the performance after the workstation is connected to the network and the selection of parallel data.

[0005] To achieve the above objectives, the present invention provides a method for constructing a plug-and-play high-performance GPU computing workstation, comprising:

[0006] Obtaining the supporting hardware type of the plug-and-play GPU, querying the version range of the supporting hardware type, and using the version range to determine the supporting hardware corresponding to the supporting hardware type;

[0007] Obtaining system software between the plug-and-play GPU and the supporting hardware, identifying parallelizable hardware in the supporting hardware based on the system software, simulating a data flow between the plug-and-play GPU and the parallelizable hardware, and analyzing the parallel performance of the parallelizable hardware using the data flow;

[0008] Obtaining a construction budget of the plug-and-play GPU, calculating a hardware budget of the parallelizable hardware, determining final hardware of the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connecting the plug-and-play GPU and the final hardware to obtain a connected workstation;

[0009] querying a construction area of ​​the connectivity workstation, simulating a path connection relationship between the connectivity workstation and a regional network within the construction area, calculating a path cost of the path connection relationship, and determining a final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path cost;

[0010] Obtaining the identity of the builder of the connectivity workstation, analyzing the type of data to be processed by the connectivity workstation using the builder identity, generating test data for the connectivity workstation using the type of data to be processed, and constructing a parallel correlation module for the connectivity workstation;

[0011] Based on the test data, data testing is performed on the connected workstation, the final connection path and the parallel correlation module to obtain data test results. According to the data test results, the connected workstation, the final connection path and the parallel correlation module are used to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

[0012] Optionally, identifying parallelizable hardware in the supporting hardware according to the system software includes:

[0013] Obtaining a software processing mode of the system software;

[0014] Check whether there is a parallel mode in the software processing method;

[0015] When a parallel mode exists in the software processing mode, identifying a hardware category related to the parallel mode;

[0016] Based on the related hardware categories, parallelizable hardware in the supporting hardware is identified.

[0017] Optionally, analyzing the parallel performance of the parallel hardware by using the data flow includes:

[0018] querying the parallel throughput of the parallel hardware for the data stream;

[0019] constructing a data volume-throughput curve between the parallel data volume of the data flow and the parallel throughput;

[0020] Using the data volume-throughput curve, determining whether the parallel data volume and the parallel throughput are correlated, and obtaining a correlation determination result;

[0021] Query the waiting time of the data flow;

[0022] The parallel performance of the parallel hardware is analyzed using the relevant judgment result and the waiting time.

[0023] Optionally, calculating the hardware budget of the parallelizable hardware includes:

[0024] Querying the construction technicians of the parallelizable hardware;

[0025] Collect the hardware cost of the parallel hardware, the construction cost of the construction technician, and the construction time cost of the parallel hardware;

[0026] The hardware budget of the parallel hardware is calculated using the hardware cost, the construction cost, and the construction time cost.

[0027] Optionally, simulating the path connection relationship between the connected workstation and the regional network within the established area includes:

[0028] Acquire network nodes in the regional network;

[0029] sending a connection request from the connectivity workstation to the network node;

[0030] Determining whether request response data from the network node is received within a preset timer period;

[0031] When request response data from the network node is received within a preset timer period, establishing a first path connection relationship between the connectivity workstation and the network node;

[0032] Acquiring a second path connection relationship of the network node in the regional network;

[0033] The first path connection relationship and the second path connection relationship are used to generate a path connection relationship between the connectivity workstation and the regional network within the construction area.

[0034] Optionally, the calculating the path cost of the path connection relationship includes:

[0035] Obtaining the shortest path between every two vertices in the path connection relationship;

[0036] Based on the shortest path, the importance of the vertices corresponding to the path connection relationship is calculated using the following formula:

[0037]

[0038] in, Represents a vertex The vertex importance, r ab (v) indicates the number of cells from the vertex To the top Passing through the vertex The number of paths, Indicates that from the vertex To the top The number of shortest paths;

[0039] According to the vertex importance, the path cost of the path connection relationship is calculated using the following formula:

[0040]

[0041] in, represents the path cost, Represents a vertex The vertex importance of Represents the distance from the connected workstation to the vertex The number of edges on a path of Indicates the number of vertices in the path connection relationship.

[0042] Optionally, generating the test data of the connected workstation by using the data type to be processed includes:

[0043] Obtaining a target data type from the data types to be processed;

[0044] Querying the input data type and output data type belonging to the target data type in the data type to be processed;

[0045] Acquire a first target type and a second target type in the target data type;

[0046] When the output data type of the first target type is consistent with the input data type of the second target type, test data of the connected workstation is generated using the output data type of the first target type and the input data type of the second target type.

[0047] Optionally, the constructing of the parallel correlation module for the connected workstations includes:

[0048] collecting input data and output data of the connected workstations;

[0049] The state transition probability between the input data and the output data is calculated using the following formula:

[0050]

[0051] in, Represents input data With output data The state transition probability between Represents input data With output data The probability of simultaneous existence, Represents input data Probability of existence;

[0052] According to the state transition probability, the state transition matrix between the input data and the output data is constructed using the following formula:

[0053]

[0054] in, represents the state transition matrix, Indicates the state transition matrix state transition probability;

[0055] Based on the state transition matrix, a trajectory analysis module of the connected workstation is determined; and an input-output correlation between the input data and the output data is calculated:

[0056]

[0057] in, represents the input-output dependency, Indicates the input data Number, Indicates the first Number, Indicates the number of input data;

[0058] Determining a correlation module of the connected workstations using the input-output correlation;

[0059] A parallel correlation module of the connected workstations is constructed by using the trajectory analysis module and the correlation module.

[0060] Optionally, performing data testing on the connected workstations, the final connection path, and the parallel correlation module based on the test data to obtain data test results includes:

[0061] inputting the test data into the communication workstation;

[0062] Under the action of the connection workstation, the final connection path and the parallel correlation module, the test data is processed to obtain a data processing status;

[0063] The speedup ratio of the connected workstations is calculated using the following formula:

[0064]

[0065] in, represents the speedup ratio, Indicates the execution time of the serial code, Indicates the execution time of the parallel code, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module;

[0066] The following formula is used to calculate the number of parallel processes of the connected workstations:

[0067]

[0068] in, Indicates the number of parallel processes. Indicates the number of parallel processes corresponding to the test data, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module;

[0069] The speedup ratio and the number of parallel processes are used as data test results.

[0070] In order to solve the above problems, the present invention also provides a plug-and-play high-performance GPU computing workstation construction system, the system comprising:

[0071] A hardware determination module is used to obtain the supporting hardware type of the plug-and-play GPU, query the version range of the supporting hardware type, and use the version range to determine the supporting hardware corresponding to the supporting hardware type;

[0072] a performance analysis module configured to obtain system software between the plug-and-play GPU and the supporting hardware, identify parallelizable hardware in the supporting hardware based on the system software, simulate a data flow between the plug-and-play GPU and the parallelizable hardware, and analyze the parallel performance of the parallelizable hardware using the data flow;

[0073] a hardware connection module, configured to obtain a construction budget of the plug-and-play GPU, calculate a hardware budget of the parallelizable hardware, determine final hardware of the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connect the plug-and-play GPU with the final hardware to obtain a connection workstation;

[0074] a path determination module, configured to query the installation area of ​​the connectivity workstation, simulate a path connection relationship between the connectivity workstation and a regional network within the installation area, calculate a path cost of the path connection relationship, and determine a final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path cost;

[0075] a module construction module, configured to obtain the identity of the builder of the connectivity workstation, analyze the type of data to be processed of the connectivity workstation using the builder identity, generate test data for the connectivity workstation using the type of data to be processed, and construct a parallel correlation module of the connectivity workstation;

[0076] A workstation construction module is used to perform data testing on the connected workstation, the final connection path and the parallel correlation module based on the test data to obtain data test results; and according to the data test results, use the connected workstation, the final connection path and the parallel correlation module to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

[0077] Compared to the problem described in the background art, the embodiment of the present invention calculates the path consumption of the path connection relationship to consider the path connection cost. Furthermore, the embodiment of the present invention constructs a parallel correlation module for the connected workstation to use the trajectory prediction method to predict the data output corresponding to each input data, and uses the correlation between each input data and the predicted output data to determine whether the input data affects the generation of the output data. If so, the input data for calculating the correlation and the input data for generating the output data cannot be parallel, otherwise they can be parallel. Therefore, the plug-and-play high-performance GPU computing workstation construction method proposed by the present invention can take into account the performance of the workstation after connecting to the network and the selection of parallel data. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A flowchart of a method for constructing a plug-and-play high-performance GPU computing workstation provided by one embodiment of the present invention;

[0079] Figure 2 A schematic diagram of modules for implementing the method for constructing a plug-and-play high-performance GPU computing workstation provided by one embodiment of the present invention.

[0080] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0081] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0082] The embodiments of the present application provide a method for constructing a plug-and-play high-performance GPU computing workstation. The execution subject of the plug-and-play high-performance GPU computing workstation construction method includes, but is not limited to, at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiments of the present application. In other words, the plug-and-play high-performance GPU computing workstation construction method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0083] Example 1:

[0084] Reference Figure 1 FIG. 1 is a flow chart of a method for constructing a plug-and-play high-performance GPU computing workstation according to an embodiment of the present invention. In this embodiment, the method for constructing a plug-and-play high-performance GPU computing workstation includes:

[0085] S1. Obtain the supporting hardware type of the plug-and-play GPU, query the version range of the supporting hardware type, and use the version range to determine the supporting hardware corresponding to the supporting hardware type.

[0086] In an embodiment of the present invention, the plug-and-play GPU refers to a graphics processing unit, the supporting hardware type refers to the type of hardware connected to the plug-and-play GPU, such as a data processing type or a data storage type, the version range refers to the version of the hardware required for data processing when a user builds a workstation, and the supporting hardware refers to hardware that meets the version range.

[0087] S2. Obtain system software between the plug-and-play GPU and the supporting hardware, identify parallel hardware in the supporting hardware based on the system software, simulate a data flow between the plug-and-play GPU and the parallel hardware, and analyze the parallel performance of the parallel hardware using the data flow.

[0088] The system software refers to an operating system, and the parallel hardware refers to hardware that can perform data parallel processing.

[0089] In one embodiment of the present invention, identifying the parallel hardware in the supporting hardware based on the system software includes: obtaining the software processing mode of the system software; querying whether there is a parallel mode in the software processing mode; when there is a parallel mode in the software processing mode, identifying the relevant hardware category of the parallel mode; based on the relevant hardware category, identifying the parallel hardware in the supporting hardware.

[0090] Optionally, the software processing method is, for example, a DMA method, in which the CPU and peripherals can operate in parallel.

[0091] Optionally, the process of simulating the data flow between the plug-and-play GPU and the parallel hardware is a process of inputting all possible input data into the parallel hardware.

[0092] In one embodiment of the present invention, the use of the data stream to analyze the parallel performance of the parallel hardware includes: querying the parallel throughput of the parallel hardware for the data stream; constructing a data volume-throughput curve between the parallel data volume of the data stream and the parallel throughput; using the data volume-throughput curve to determine whether the parallel data volume and the parallel throughput are correlated, and obtaining a correlation judgment result; querying the waiting time of the data stream; and using the correlation judgment result and the waiting time to analyze the parallel performance of the parallel hardware.

[0093] The waiting time refers to the time that the data stream waits to be called, and the parallel performance includes the relevant judgment result and the waiting time.

[0094] S3. Obtain a construction budget for the plug-and-play GPU, calculate a hardware budget for the parallelizable hardware, determine final hardware for the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connect the plug-and-play GPU to the final hardware to obtain a connected workstation.

[0095] The construction budget refers to the budget for building a workstation, and the hardware budget refers to the actual construction cost.

[0096] In one embodiment of the present invention, the calculation of the hardware budget of the parallel hardware includes: querying the construction technician of the parallel hardware; collecting the hardware cost of the parallel hardware, the construction cost of the construction technician and the construction time cost of the parallel hardware; and using the hardware cost, the construction cost and the construction time cost to calculate the hardware budget of the parallel hardware.

[0097] The construction time cost refers to the cost loss caused by the lack of workstations during the waiting time for the construction to be completed.

[0098] Optionally, the process of determining the final hardware of the plug-and-play GPU from the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget refers to selecting several batches of hardware combinations with good parallel performance and hardware budgets lower than the construction budget.

[0099] S4. Query the construction area of ​​the connectivity workstation, simulate the path connection relationship between the connectivity workstation and the regional network within the construction area, calculate the path consumption of the path connection relationship, and determine the final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path consumption.

[0100] The construction area refers to the area where the workstation is constructed.

[0101] In one embodiment of the present invention, the simulation of the path connection relationship between the connectivity workstation and the regional network in the construction area includes: obtaining a network node in the regional network; sending a connection request from the connectivity workstation to the network node; determining whether request response data from the network node is received within a preset timer period; when the request response data from the network node is received within the preset timer period, constructing a first path connection relationship between the connectivity workstation and the network node; obtaining a second path connection relationship of the network node in the regional network; and using the first path connection relationship and the second path connection relationship to generate a path connection relationship between the connectivity workstation and the regional network in the construction area.

[0102] The network node refers to other devices that can establish a connection with the connected workstation, and the second path connection relationship refers to the connection relationship between other nodes and the network node.

[0103] Furthermore, the embodiment of the present invention calculates the path consumption of the path connection relationship to consider the path connection cost.

[0104] In one embodiment of the present invention, calculating the path cost of the path connection relationship includes: obtaining the shortest path between every two vertices in the path connection relationship; and calculating the importance of the vertex corresponding to the path connection relationship based on the shortest path using the following formula:

[0105]

[0106] in, Represents a vertex The vertex importance, r ab (v) indicates the number of cells from the vertex To the top Passing through the vertex The number of paths, Indicates that from the vertex To the top The number of shortest paths;

[0107] According to the vertex importance, the path cost of the path connection relationship is calculated using the following formula:

[0108]

[0109] in, represents the path cost, Represents a vertex The vertex importance of Represents the distance from the connected workstation to the vertex The number of edges on a path of Indicates the number of vertices in the path connection relationship.

[0110] S5. Obtain the identity of the builder of the connectivity workstation, use the builder identity to analyze the type of data to be processed of the connectivity workstation, use the type of data to be processed to generate test data of the connectivity workstation, and construct a parallel correlation module of the connectivity workstation.

[0111] The establishment identity refers to an identity that includes the type of data to be processed. For example, if the demander of the workstation is a hospital, the type of data to be processed is a medical data type.

[0112] In one embodiment of the present invention, the use of the data type to be processed to generate test data for the connectivity workstation includes: obtaining a target data type in the data type to be processed; querying the input data type and output data type belonging to the target data type in the data type to be processed; obtaining a first target type and a second target type in the target data type; and when the output data type of the first target type is consistent with the input data type of the second target type, generating test data for the connectivity workstation using the output data type of the first target type and the input data type of the second target type.

[0113] Among them, the target data type refers to any type of data type to be processed, the input data type and the output data type are respectively the data that can obtain the target data type through calculation, and the result that can be obtained by calculating the target data type, and the first target type and the second target type are any two different data types of the target data type.

[0114] Furthermore, an embodiment of the present invention constructs a parallel correlation module of the connected workstation to predict the data output corresponding to each input data using a trajectory prediction method, and uses the correlation between each input data and the predicted output data to determine whether the input data affects the generation of the output data. If so, the input data for calculating the correlation and the input data for generating the output data cannot be calculated in parallel, otherwise they can be calculated in parallel.

[0115] In one embodiment of the present invention, the parallel correlation module for constructing the connected workstation includes: collecting input data and output data of the connected workstation; and calculating the state transition probability between the input data and the output data using the following formula:

[0116]

[0117] in, Represents input data With output data The state transition probability between Represents input data With output data The probability of simultaneous existence, Represents input data Probability of existence;

[0118] According to the state transition probability, the state transition matrix between the input data and the output data is constructed using the following formula:

[0119]

[0120] in, represents the state transition matrix, Indicates the state transition matrix state transition probability;

[0121] Based on the state transition matrix, a trajectory analysis module of the connected workstation is determined; and an input-output correlation between the input data and the output data is calculated:

[0122]

[0123] in, represents the input-output dependency, Indicates the input data Number, Indicates the first Number, Indicates the number of input data;

[0124] The input-output correlation is used to determine a correlation module of the connected workstation; and a parallel correlation module of the connected workstation is constructed by using the trajectory analysis module and the correlation module.

[0125] The trajectory analysis module and the correlation module refer to computing systems that can calculate state transition probabilities and correlations.

[0126] S5. Based on the test data, data testing is performed on the connected workstation, the final connection path, and the parallel correlation module to obtain data test results. According to the data test results, the connected workstation, the final connection path, and the parallel correlation module are used to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

[0127] In one embodiment of the present invention, performing data testing on the connectivity workstation, the final connection path, and the parallel correlation module based on the test data to obtain data test results includes: inputting the test data into the connectivity workstation; processing the test data under the control of the connectivity workstation, the final connection path, and the parallel correlation module to obtain data processing status; and calculating the speedup ratio of the connectivity workstation using the following formula:

[0128]

[0129] in, represents the speedup ratio, Indicates the execution time of the serial code, Indicates the execution time of the parallel code, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module;

[0130] The following formula is used to calculate the number of parallel processes of the connected workstations:

[0131]

[0132] in, Indicates the number of parallel processes. Indicates the number of parallel processes corresponding to the test data, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module;

[0133] The speedup ratio and the number of parallel processes are used as data test results.

[0134] It should be noted that the execution time and the number of processors are obtained from the data processing situation.

[0135] Optionally, the process of completing the workstation construction of the plug-and-play GPU using the connected workstation, the final connection path and the parallel correlation module based on the data test results to obtain the workstation construction result of the plug-and-play GPU refers to using the connected workstation, the final connection path and the parallel correlation module as the workstation construction result of the plug-and-play GPU when the data test results perform well.

[0136] Compared to the problems described in the background technology, the embodiments of the present invention calculate the path consumption of the path connection relationship to consider the path connection cost. Furthermore, the embodiments of the present invention construct a parallel correlation module of the connected workstation to predict the data output corresponding to each input data using the trajectory prediction method, and use the correlation between each input data and the predicted output data to determine whether the input data affects the generation of the output data. If so, the input data for calculating the correlation and the input data for generating the output data cannot be parallel, otherwise they can be parallel. Therefore, the plug-and-play high-performance GPU computing workstation construction method proposed by the present invention can take into account the performance after the workstation is connected to the network and the selection of parallel data.

[0137] Example 2:

[0138] like Figure 2 The figure shows a functional module diagram of a plug-and-play high-performance GPU computing workstation construction system of the present invention.

[0139] The plug-and-play high-performance GPU computing workstation construction system 200 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the plug-and-play high-performance GPU computing workstation construction system may include a hardware determination module 201, a performance analysis module 202, a hardware connectivity module 203, a path determination module 204, a module construction module 205, and a workstation construction module 206. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0140] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0141] The hardware determination module 201 is configured to obtain the supporting hardware type of the plug-and-play GPU, query the version range of the supporting hardware type, and determine the supporting hardware corresponding to the supporting hardware type using the version range;

[0142] The performance analysis module 202 is configured to obtain system software between the plug-and-play GPU and the supporting hardware, identify parallelizable hardware in the supporting hardware based on the system software, simulate a data flow between the plug-and-play GPU and the parallelizable hardware, and analyze the parallel performance of the parallelizable hardware using the data flow;

[0143] The hardware connection module 203 is configured to obtain a construction budget of the plug-and-play GPU, calculate a hardware budget of the parallelizable hardware, determine the final hardware of the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connect the plug-and-play GPU with the final hardware to obtain a connected workstation;

[0144] The path determination module 204 is configured to query the installation area of ​​the connectivity workstation, simulate a path connection relationship between the connectivity workstation and a regional network within the installation area, calculate a path cost of the path connection relationship, and determine a final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path cost;

[0145] The module construction module 205 is configured to obtain the identity of the builder of the connectivity workstation, analyze the data type to be processed of the connectivity workstation using the builder identity, generate test data for the connectivity workstation using the data type to be processed, and construct a parallel correlation module for the connectivity workstation;

[0146] The workstation construction module 206 is used to perform data testing on the connected workstation, the final connection path and the parallel correlation module based on the test data to obtain data test results. According to the data test results, the connected workstation, the final connection path and the parallel correlation module are used to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

[0147] In detail, the modules in the plug-and-play high-performance GPU computing workstation construction system 200 in the embodiment of the present invention are used in the same manner as above. Figure 1 The technical means are the same as the plug-and-play high-performance GPU computing workstation construction method described in and can produce the same technical effects, so I will not go into details here.

[0148] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a plug-and-play high-performance GPU computing workstation, characterized in that: The method comprises: Obtaining the supporting hardware type of the plug-and-play GPU, querying the version range of the supporting hardware type, and using the version range to determine the supporting hardware corresponding to the supporting hardware type; Obtaining system software between the plug-and-play GPU and the supporting hardware, identifying parallelizable hardware in the supporting hardware based on the system software, simulating a data flow between the plug-and-play GPU and the parallelizable hardware, and analyzing the parallel performance of the parallelizable hardware using the data flow; Obtaining a construction budget of the plug-and-play GPU, calculating a hardware budget of the parallelizable hardware, determining final hardware of the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connecting the plug-and-play GPU and the final hardware to obtain a connected workstation; querying a construction area of ​​the connectivity workstation, simulating a path connection relationship between the connectivity workstation and a regional network within the construction area, calculating a path cost of the path connection relationship, and determining a final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path cost; Obtaining the identity of the builder of the connectivity workstation, analyzing the type of data to be processed by the connectivity workstation using the builder identity, generating test data for the connectivity workstation using the type of data to be processed, and constructing a parallel correlation module for the connectivity workstation; Based on the test data, data testing is performed on the connected workstation, the final connection path and the parallel correlation module to obtain data test results. According to the data test results, the connected workstation, the final connection path and the parallel correlation module are used to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

2. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The step of identifying parallelizable hardware in the supporting hardware according to the system software includes: Obtaining a software processing mode of the system software; Check whether there is a parallel mode in the software processing method; When a parallel mode exists in the software processing mode, identifying a hardware category related to the parallel mode; Based on the related hardware categories, parallelizable hardware in the supporting hardware is identified.

3. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The analyzing the parallel performance of the parallel hardware by using the data stream includes: querying the parallel throughput of the parallel hardware for the data stream; constructing a data volume-throughput curve between the parallel data volume of the data flow and the parallel throughput; Using the data volume-throughput curve, determining whether the parallel data volume and the parallel throughput are correlated, and obtaining a correlation determination result; Query the waiting time of the data flow; The parallel performance of the parallel hardware is analyzed using the relevant judgment result and the waiting time.

4. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: Calculating the hardware budget of the parallelizable hardware includes: Querying the construction technicians of the parallelizable hardware; Collect the hardware cost of the parallel hardware, the construction cost of the construction technician, and the construction time cost of the parallel hardware; The hardware budget of the parallel hardware is calculated using the hardware cost, the construction cost, and the construction time cost.

5. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The simulating the path connection relationship between the connected workstation and the regional network within the established area includes: Acquire network nodes in the regional network; sending a connection request from the connectivity workstation to the network node; Determining whether request response data from the network node is received within a preset timer period; When request response data from the network node is received within a preset timer period, establishing a first path connection relationship between the connectivity workstation and the network node; Acquiring a second path connection relationship of the network node in the regional network; The first path connection relationship and the second path connection relationship are used to generate a path connection relationship between the connectivity workstation and the regional network within the construction area.

6. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The calculating the path cost of the path connection relationship includes: Obtaining the shortest path between every two vertices in the path connection relationship; Based on the shortest path, the importance of the vertices corresponding to the path connection relationship is calculated using the following formula: in, Represents a vertex The vertex importance, r ab (v) indicates the number of cells from the vertex To the top Passing through the vertex The number of paths, Indicates that from the vertex To the top The number of shortest paths; According to the vertex importance, the path cost of the path connection relationship is calculated using the following formula: in, represents the path cost, Represents a vertex The vertex importance of Represents the distance from the connected workstation to the vertex The number of edges on a path of Indicates the number of vertices in the path connection relationship.

7. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The generating the test data of the connected workstation by using the data type to be processed includes: Obtaining a target data type from the data types to be processed; Querying the input data type and output data type belonging to the target data type in the data type to be processed; Acquire a first target type and a second target type in the target data type; When the output data type of the first target type is consistent with the input data type of the second target type, test data of the connected workstation is generated using the output data type of the first target type and the input data type of the second target type.

8. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The parallel dependency module for constructing the connected workstations includes: collecting input data and output data of the connected workstations; The state transition probability between the input data and the output data is calculated using the following formula: in, Represents input data With output data The state transition probability between Represents input data With output data The probability of simultaneous existence, Represents input data Probability of existence; According to the state transition probability, the state transition matrix between the input data and the output data is constructed using the following formula: in, represents the state transition matrix, Indicates the state transition matrix state transition probability; Based on the state transition matrix, a trajectory analysis module of the connected workstation is determined; and an input-output correlation between the input data and the output data is calculated: in, represents the input-output dependency, Indicates the input data Number, Indicates the first Number, Indicates the number of input data; Determining a correlation module of the connected workstations using the input-output correlation; A parallel correlation module of the connected workstations is constructed by using the trajectory analysis module and the correlation module.

9. The method for constructing a plug-and-play high-performance GPU computing workstation according to claim 1, wherein: The step of performing data testing on the connected workstation, the final connection path, and the parallel correlation module based on the test data to obtain data test results includes: inputting the test data into the communication workstation; Under the action of the connection workstation, the final connection path and the parallel correlation module, the test data is processed to obtain a data processing status; The speedup ratio of the connected workstations is calculated using the following formula: in, represents the speedup ratio, Indicates the execution time of the serial code, Indicates the execution time of the parallel code, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module; The following formula is used to calculate the number of parallel processes of the connected workstations: in, Indicates the number of parallel processes. Indicates the number of parallel processes corresponding to the test data, represents the number of processors under the action of the connected workstation, the final connection path and the parallel dependency module; The speedup ratio and the number of parallel processes are used as data test results.

10. A plug-and-play high-performance GPU computing workstation construction system, characterized in that: The system comprises: A hardware determination module is used to obtain the supporting hardware type of the plug-and-play GPU, query the version range of the supporting hardware type, and use the version range to determine the supporting hardware corresponding to the supporting hardware type; a performance analysis module configured to obtain system software between the plug-and-play GPU and the supporting hardware, identify parallelizable hardware in the supporting hardware based on the system software, simulate a data flow between the plug-and-play GPU and the parallelizable hardware, and analyze the parallel performance of the parallelizable hardware using the data flow; a hardware connection module, configured to obtain a construction budget of the plug-and-play GPU, calculate a hardware budget of the parallelizable hardware, determine final hardware of the plug-and-play GPU in the parallelizable hardware using the parallel performance, the construction budget, and the hardware budget, and connect the plug-and-play GPU with the final hardware to obtain a connection workstation; a path determination module, configured to query the installation area of ​​the connectivity workstation, simulate a path connection relationship between the connectivity workstation and a regional network within the installation area, calculate a path cost of the path connection relationship, and determine a final connection path between the connectivity workstation and the regional network in the path connection relationship based on the path cost; a module construction module, configured to obtain the identity of the builder of the connectivity workstation, analyze the type of data to be processed of the connectivity workstation using the builder identity, generate test data for the connectivity workstation using the type of data to be processed, and construct a parallel correlation module of the connectivity workstation; A workstation construction module is used to perform data testing on the connected workstation, the final connection path and the parallel correlation module based on the test data to obtain data test results; and according to the data test results, use the connected workstation, the final connection path and the parallel correlation module to complete the workstation construction of the plug-and-play GPU to obtain the workstation construction result of the plug-and-play GPU.

Images