Heterogeneous intelligent computing chip RDMA communication method, system and platform and storage medium

Through the RDMA communication method, the problem of interconnection between heterogeneous intelligent computing chips is solved, the effective utilization of high-performance communication and computing resources is realized, the application scenarios are expanded, and the poor communication performance and low expansion capabilities in the existing technology are solved. question.

CN119046210AActive Publication Date: 2024-11-29SHANG HAI JIE YUE XING CHEN ZHI NENG KE JI YOU XIAN GONG SI
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411130969.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-29
Estimated Expiration
2044-08-16

AI Technical Summary

Technical Problem

The existing technology is difficult to achieve efficient interconnection between different intelligent computing chips, resulting in low utilization of computing resources and limited application scenarios, especially in the fields of large model training and inference, there are problems such as poor communication performance and low scaling capabilities.

Method used

The RDMA communication method is adopted to ensure communication performance and resource utilization by identifying and initializing heterogeneous intelligent computing chips, establishing in-machine interconnection topology, configuring network interfaces, creating RDMA communication channels, and performing data transmission and abnormality detection processing.

Benefits of technology

It realizes high-performance communication between heterogeneous intelligent computing chips, improves computing resource utilization, expands application scenarios, solves the problem of interconnection between different chips, and provides better computing resource management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119046210A_ABST
    Figure CN119046210A_ABST
Patent Text Reader

Abstract

The invention discloses a heterogeneous intelligent computing chip RDMA communication method, system and platform and a storage medium, and the communication method comprises the steps: executing RDMA communication initialization, so as to carry out intelligent computing chip recognition and built-in interconnection topology discovery; performing RDMA communication data transmission on to-be-processed data based on the identified intelligent computing chip and the discovered built-in interconnection topology; executing an RDMA communication exit operation so as to release the occupied intelligent computing chip resources; and performing RDMA communication anomaly detection and processing operation to identify whether a communication anomaly event occurs and process the identified communication anomaly event. According to the heterogeneous intelligent computing chip RDMA communication method provided by the invention, the communication performance can be ensured by adopting the heterogeneous intelligent computing chip hybrid communication-oriented RDMA communication method, so that the utilization rate of computing resources is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a heterogeneous intelligent computing chip RDMA communication method, system, heterogeneous intelligent computing chip computing platform and storage medium. Background Art

[0002] The large-scale parameters in large model training put forward higher requirements for computing power and video memory. Taking GPT3 as an example, 100 billion parameters require 2TB of video memory, and the current single-card video memory capacity is not enough. In order to shorten the training time, distributed training technology is usually used to split the model and data, and the training time is shortened by using multiple machines and multiple cards. However, the current intelligent computing chips designed and produced by different intelligent computing chip manufacturers use different chip architectures, software ecosystems and communication protocols, which are incompatible with each other and cannot be interconnected at high speed, making it difficult to form a synergy in the field of large model training / inference that requires ultra-large computing power resources. For example: BR100 of Biren Technology, BI-V100 of Tianshu Zhixin and DCU series chips of Haiguang Technology are based on the general GPU architecture, but the software of the former two is compatible with Nvidia CUDA ecosystem, while Haiguang DCU supports AMD ROCm ecosystem. Unlike the above manufacturers, Huawei NPU910B adopts DSA architecture, adopts its own software stack, and does not support CUDA and ROCm ecosystems. Huawei NPU board integrates high-performance network interface to implement private communication protocol. Therefore, how to overcome the interconnection and interoperability issues between different intelligent computing chips and realize hybrid communication between heterogeneous intelligent computing chips at home and abroad is an important technical challenge with extremely high economic value.

[0003] At present, many domestic and foreign institutions and manufacturers have begun to pay attention to the interoperability issues between different intelligent computing chips and conduct research. However, the current mainstream solutions mainly focus on the discussion and research of mixed networking training of chips from the same manufacturer / software ecosystem / communication protocol, and do not consider the mixed communication issues of heterogeneous intelligent computing chips. How to solve the problem of mixed communication of heterogeneous intelligent computing chips is a technical problem that needs to be solved urgently in this field.

[0004] In an existing artificial intelligence (AI) cluster solution that supports multi-core hybrid training, a low-speed TCP (Transmission Control Protocol) method is used to solve the problem of cross-protocol communication of heterogeneous chips. However, this solution has disadvantages such as poor communication performance and low scalability, which makes it difficult to achieve the optimal computing resource utilization and limited application scenarios. Therefore, it is necessary to propose a high-performance communication method for heterogeneous intelligent computing chips to ensure communication performance and thus improve computing resource utilization. Summary of the invention

[0005] To this end, the present invention provides a heterogeneous intelligent computing chip RDMA communication method, system, platform and storage medium, in an effort to solve or at least alleviate at least one of the above problems.

[0006] According to one aspect of the present invention, a heterogeneous intelligent computing chip RDMA communication method is provided, the method comprising:

[0007] Step S100: Execute RDMA communication initialization to identify the intelligent computing chip and discover the internal interconnection topology;

[0008] Step S200: performing RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology;

[0009] Step S300: Execute the RDMA communication exit operation to release the occupied intelligent computing chip resources;

[0010] Step S400: performing RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and process the identified communication anomaly event.

[0011] Exemplarily, the step S100 includes:

[0012] Step S101: performing an intelligent computing chip discovery operation to identify and discover available intelligent computing chips in the system;

[0013] Step S102: performing an intra-machine interconnection topology discovery operation to determine the interconnection topology relationship between the discovered intelligent computing chips;

[0014] Step S103: performing a network interface discovery operation to identify a network interface that can be used for RDMA communication;

[0015] Step S104: Execute RDMA resource initialization operation;

[0016] Step S105: performing synchronization operations on the RDMA resource information and the interconnection information.

[0017] Exemplarily, step S101 is implemented by polling PCIe, including the following steps:

[0018] Step S1011: Initialize the PCIe polling mechanism to periodically read PCIe device information;

[0019] Step S1012: performing a PCIe polling operation, and reading PCIe device information in each polling cycle, wherein the PCIe device information includes a device list and configuration information;

[0020] Step S1013: Analyze the read PCIe device information to determine whether there is an intelligent computing chip.

[0021] Optionally, the step S102 includes:

[0022] Step S1021: performing a connection relationship detection operation to determine the physical connection mode between the intelligent computing chips in the server;

[0023] Step S1022: constructing an interconnection topology structure inside the server based on the physical connection method.

[0024] Optionally, the step S103 includes:

[0025] Step S1031: performing a protocol identification operation to identify a network protocol supporting RDMA in the system;

[0026] Step S1032: searching for devices supporting RDMA in the network according to the identified network protocol;

[0027] Step S1033: Acquire the configuration information of the device supporting RDMA, wherein the configuration information includes an IP address and a port number.

[0028] Optionally, the step S104 includes:

[0029] Step S1041: performing RDMA initialization on the intelligent computing chip to ensure that it works normally and supports RDMA communication;

[0030] Step S1042: creating an RDMA communication channel for the intelligent computing chip that has completed RDMA initialization;

[0031] Step S1043: performing parameter configuration on the created RDMA communication channel, wherein the configuration includes the peer QP number and the peer IP address;

[0032] Step S1044: Execute a memory area registration operation to register the memory area of ​​the application program in the intelligent computing chip, so that the intelligent computing chip can directly access the memory area for data transmission.

[0033] Optionally, the step S200 includes:

[0034] Receive and identify the PrepareToRecv signal from the intelligent computing chip as the receiving end;

[0035] Sending the data to be processed to the intelligent computing chip based on the RDMA WRITE operation, so as to store the data to be processed into the memory of the receiving end;

[0036] After the data to be processed is sent, a SendDone signal is sent to the intelligent computing chip;

[0037] Receive and identify the SendDoneAck signal from the intelligent computing chip to end data transmission.

[0038] Optionally, in step S200, a memory dedicated to receiving signaling is set for the receiving end.

[0039] Optionally, the step S300 includes:

[0040] Step S301: closing the physical connection and logical connection related to the RDMA communication;

[0041] Step S302: releasing memory resources used in the RDMA communication process;

[0042] Step S303: canceling the registration of the memory area of ​​the intelligent computing chip during the RDMA communication process;

[0043] Step S304: clearing the status information related to the RDMA communication.

[0044] Exemplarily, the step S400 includes:

[0045] In step S100, initialization anomaly detection and processing are performed, wherein the initialization anomaly includes at least one of RDMA network connectivity anomaly, device discovery anomaly, configuration error, and insufficient resources; and / or,

[0046] In the step S200, data transmission anomaly detection and processing are performed, wherein the data transmission anomaly includes at least one of abnormal packet loss, link interruption, congestion control, and data consistency; and / or,

[0047] In the step S300, communication exit anomaly detection and processing is performed, wherein the communication exit anomaly includes at least one of resource release anomaly, connection closing anomaly, and configuration recovery anomaly.

[0048] Another embodiment of the present invention provides a heterogeneous intelligent computing chip RDMA communication system, the system comprising:

[0049] Heterogeneous intelligent computing chip RDMA communication initialization unit, used to perform RDMA communication initialization to identify intelligent computing chips and discover the internal interconnection topology;

[0050] A heterogeneous intelligent computing chip RDMA communication data transmission unit is used to perform RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology;

[0051] A heterogeneous intelligent computing chip RDMA communication exit operation unit, used to perform an RDMA communication exit operation to release the occupied intelligent computing chip resources;

[0052] The heterogeneous intelligent computing chip RDMA communication anomaly detection and processing unit is used to perform RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and to process the identified communication anomaly event.

[0053] Another aspect of the present invention also provides a heterogeneous intelligent computing chip computing platform, which includes at least two heterogeneous intelligent computing chips, wherein data communication in the heterogeneous intelligent computing chip platform adopts the heterogeneous intelligent computing chip RDMA communication method described in any one of the above items.

[0054] Another aspect of the present invention provides a readable storage medium storing program instructions. When the program instructions are read and executed by a computing device, the computing device executes any one of the heterogeneous intelligent computing chip RDMA communication methods described above.

[0055] According to the heterogeneous intelligent computing chip RDMA communication method of the present invention, by adopting the RDMA communication method for hybrid communication of heterogeneous intelligent computing chips, the communication performance can be guaranteed, thereby improving the utilization rate of computing resources. Correspondingly, the heterogeneous intelligent computing chip RDMA communication system, heterogeneous intelligent computing chip computing platform and storage medium of the present invention also have the aforementioned technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings. The drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the drawings, the same reference numerals generally represent the same components or steps.

[0057] Figure 1 It is a schematic flow chart of a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention;

[0058] Figure 2 is a schematic block diagram of an RDMA communication initialization method according to an embodiment of the present invention;

[0059] Figure 3 is a schematic flow chart of a method for identifying and discovering an intelligent computing chip by polling PCIe according to an embodiment of the present invention;

[0060] Figure 4 is a schematic flow chart of a method for discovering an intra-machine interconnection topology according to an embodiment of the present invention;

[0061] Figure 5 is a schematic flow chart of a RDMA network interface discovery method according to an embodiment of the present invention;

[0062] Figure 6 is a schematic flow chart of a RDMA resource initialization method according to an embodiment of the present invention;

[0063] Figure 7 is a schematic flow chart of an RDMA communication data transmission method according to an embodiment of the present invention;

[0064] Figure 8 It is a schematic flow chart of a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention;

[0065] Fig. 9 It is a schematic diagram of a hybrid communication SDK software system running on an intelligent computing server according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0067] In the embodiments of the present application, the intelligent computing chip refers to a chip used for artificial intelligence computing tasks, including GPU, NPU, FPGA and other chips. Among them, the artificial computing tasks covered by this application include artificial intelligence training tasks and artificial intelligence reasoning tasks. Heterogeneous intelligent computing chips refer to an artificial intelligence computing task completed by different intelligent computing chips, where the differences in intelligent computing chips include manufacturers, software ecology, chip architecture, etc. RDMA (Remote Direct Memory Access), that is, remote direct data access, is a high-performance network communication technology with the advantages of high bandwidth, low latency, and no CPU consumption. RDMA is specific to the protocol level, mainly including three protocols: Infiniband (IB), RDMA over Converged Ethernet (RoCE) and Internet Wide Area RDMA Protocol (iWARP). These three protocols all comply with the RDMA standard and share the same upper user interface (Verbs), but there are some differences at different levels. RDMA communication refers to the transmission of data through RDMA. Traditional RDMA communication is usually carried out under the same protocol. The heterogeneous intelligent computing chip RDMA communication in the embodiment of the present application refers to the communication between heterogeneous computing chips based on the RDMA protocol, which is obviously cross-protocol communication. An important inventive point of an embodiment of the present application is to propose a specific implementation method for RDMA communication of heterogeneous intelligent computing chips.

[0068] Figure 1 The overall process of a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention is shown, which includes steps S100 to S400:

[0069] Step S100: RDMA communication initialization.

[0070] This step refers to the stage of device discovery, establishing a communication connection or initializing communication parameters. In the embodiment of the present application, step S100 can be adopted Figure 2 The method shown is implemented, and other feasible methods can also be used to implement it. In this step, RDMA communication initialization is performed to identify the intelligent computing chip and discover the internal interconnection topology.

[0071] Step S200: RDMA communication data transmission.

[0072] This step is used to perform data transmission related operations, specifically: based on the identified intelligent computing chip and the discovered internal interconnection topology, RDMA communication data transmission is performed on the data to be processed. Among them, data transmission in the embodiment of the present invention refers to the communication process of transmitting data from one place to another.

[0073] Step S300: RDMA communication exits.

[0074] This step marks the end of the communication process, including closing the connection or releasing related resources. When the program completes step S200, the RDMA communication exit operation will be executed. The RDMA communication exit process is started to ensure that system resources (i.e., intelligent computing chip resources) are correctly released and the system returns to the initial state.

[0075] Step S400: RDMA communication anomaly detection and processing.

[0076] This step is used to perform RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event has occurred and to handle the identified communication anomaly event. The heterogeneous intelligent computing chip RDMA communication anomaly processing process refers to the processing process taken by the system when an abnormal situation occurs during the RDMA communication process. The purpose of this process is to ensure that the system can correctly handle abnormal situations and take appropriate measures to restore communication or minimize losses.

[0077] exist Figure 1 In the flowchart shown, the direction of the arrows indicates the execution order between these steps. There is a direct arrow between step S100 and step S200, indicating that data transmission is immediately followed by initialization. There is also a direct arrow between step S200 and step S300, indicating that communication is immediately exited after data transmission. In addition, steps S100, S200 and S300 all point to step S400, indicating that in any communication process, if an exception occurs, the exception handling step will be executed, and the original step will be returned after the processing is completed and restored.

[0078] Figure 2 FIG. 1 shows an exemplary detailed process of step S100, that is, the RDMA communication initialization step, in a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention. Figure 2 As shown, step S100 may include steps S101 to S105, which are specifically as follows:

[0079] Step S101: Intelligent computing chip device discovery.

[0080] Intelligent computing chip device discovery, also known as intelligent computing chip automatic identification and discovery, is the first step in the RDMA communication initialization process, which involves identifying and discovering intelligent computing chip devices. This step may include searching for available intelligent computing resources in the network or detecting specific devices.

[0081] In this embodiment, automatic identification and discovery of intelligent computing chips refers to the software system automatically discovering the intelligent computing chips installed in the server to achieve the management and utilization of these chips. In the prior art, servers often install intelligent computing chips and other external devices through the PCIe (Peripheral Component Interconnect Express, a high-speed serial computer expansion bus standard) interface. In the existing non-heterogeneous intelligent computing chip solutions, since the programming interfaces supported by the intelligent computing chips are the same, the intelligent computing chips can usually be automatically discovered directly through the programming interfaces during communication. However, for systems including heterogeneous intelligent computing chips, since the programming interfaces supported by the heterogeneous intelligent computing chips are quite different, automatic discovery cannot be performed directly through these interfaces. In order to solve this problem, the embodiment of the present application proposes a solution for completing the identification and discovery of intelligent computing chips by polling PCIe. By polling PCIe, the current network information is read, and information related to the intelligent computing chip is obtained therefrom.

[0082] Figure 3 FIG. 2 shows an exemplary detailed process of identifying and discovering an intelligent computing chip by polling PCIe according to an embodiment of the present invention. Figure 3 As shown, the method for identifying and discovering the intelligent computing chip by polling PCIe includes the following steps:

[0083] Step S1011: Initialize the PCIe polling mechanism.

[0084] The PCIe polling mechanism is initialized to periodically read PCIe device information.

[0085] Step S1012: Execute a polling operation to read PCIe device information in each polling cycle.

[0086] Perform polling operations and read PCIe device information in each polling cycle, including device lists, configuration information, etc.

[0087] Step S1013: Analyze the read PCIe device information to determine whether there is an intelligent computing chip.

[0088] The method for determining whether there is an intelligent computing chip may be to search for a specific device identifier, configuration parameter or other feature. If the PCIe device information includes a specific device identifier or a specific configuration parameter, it is determined that there is an intelligent computing chip on the corresponding interface.

[0089] The embodiments of the present application can realize automatic discovery of intelligent computing chips in the server by polling PCIe without relying on specific programming interfaces of heterogeneous intelligent computing chips.

[0090] Step S102: discovering the intra-machine interconnection topology.

[0091] After executing step S101 to discover the intelligent computing chip devices, the next step is to determine the internal connection structure between these intelligent computing chip devices, that is, the server internal interconnection topology. Through this step, the physical or logical connection mode between the devices can be understood for subsequent communication configuration.

[0092] In this application, server internal interconnection topology discovery refers to the automatic identification and discovery of the interconnection topology structure inside the intelligent computing server through specific methods and algorithms. In an intelligent computing server, there are usually multiple computing nodes and network devices, such as GPU cards, CPUs, memories, storage devices, etc. These devices are connected through various interconnection technologies, such as PCIe, NVLink, InfiniBand, etc. Since the interconnection methods and topological structures of these devices may be very complex, and the interconnection methods and topological structures of different intelligent computing chips are basically different, special technologies are required to automatically discover and identify them.

[0093] Figure 4 FIG. 2 shows an exemplary process of an intra-machine interconnection topology discovery method according to an embodiment of the present invention. Figure 4 As shown, the method for discovering the intra-machine interconnection topology in the embodiment of the present application includes the following steps:

[0094] Step S1021: connection detection. This step specifically includes: performing connection relationship detection to determine the physical connection mode between various devices in the server.

[0095] This step determines the physical connection method between devices in the server by analyzing the connection relationship between them, such as PCIe links, NVLink channels, etc.

[0096] Step S1022: Topology construction: This step specifically includes: constructing the interconnection topology structure inside the server based on the physical connection mode.

[0097] This step builds the interconnection topology structure inside the server according to the connection relationship and configuration information between devices, where the interconnection topology structure includes the connection mode, bandwidth, delay, etc. between nodes.

[0098] The in-machine interconnection topology discovery method of the embodiment of the present application can obtain the physical or logical connection mode information between devices through link detection and topology construction, which can be used for subsequent communication configuration.

[0099] Step S103: Network interface discovery.

[0100] The purpose of this step is to identify a network interface that can be used for RDMA communication. The specific implementation of network interface discovery can be to search for a network card that supports RDMA, to search for a port that supports RDMA, or to search for other network hardware resources that support RDMA, etc., which are not limited here.

[0101] In this application, RDMA network interface discovery refers to automatically identifying and discovering a network interface that supports RDMA through specific methods and protocols. Figure 5 FIG. 2 shows an exemplary process of an RDMA network interface discovery method according to an embodiment of the present invention. Figure 5 As shown, based on the discovery of the intra-machine topology interconnection, the RDMA network interface discovery method of the present application mainly includes the following steps:

[0102] Step S1031: protocol identification, that is, by analyzing network data packets or querying information of network devices, a network protocol supporting RDMA, such as InfiniBand, RoCE (RDMA over Converged Ethernet) or iWARP (Internet Wide Area RDMA Protocol), is identified.

[0103] Step S1032: Device discovery, that is, searching for devices supporting RDMA in the network, such as RDMA network cards (HCA or RNIC) according to the identified protocol. This can be achieved by sending a specific network request or querying device information.

[0104] Step S1033: configuration acquisition, that is, for the discovered devices supporting RDMA, their configuration information, such as IP addresses, port numbers, etc., is acquired.

[0105] Step S104: Initialize RDMA resources.

[0106] After discovering the network interface, the step to be performed is to initialize the RDMA resources. In this step, the RDMA resource initialization may include: configuring the network interface for RDMA communication, establishing a communication channel, and negotiating communication parameters.

[0107] Figure 6 FIG. 2 shows an exemplary process of an RDMA resource initialization method according to an embodiment of the present invention. Figure 6 As shown, in the present application, after the RDMA network interface discovery is completed, the method of initializing RDMA related resources based on the acquired RDMA configuration includes the following steps:

[0108] Step S1041: Initialize the RDMA device.

[0109] This step is to ensure that the device can work properly and support RDMA communication.

[0110] Specifically, you can use the API or library function provided by the manufacturer to initialize the RDMA device, including configuring device parameters, setting the working mode, etc.

[0111] Step S1042: Create an RDMA communication channel.

[0112] Based on the initialized RDMA device, create RDMA communication channels, such as Queue Pair (QP) or Completion Queue (CQ). These channels will be used for data transmission and communication control between devices.

[0113] Step S1043: Configure the RDMA communication channel.

[0114] In this step, the parameters of the created RDMA communication channel are configured. The configuration includes the peer QP number and the peer IP address. In a specific implementation, the configured communication channel parameters include the communication mode, transmission protocol, data packet size, etc. These parameters will directly affect the performance and efficiency of the communication.

[0115] Step S1044: Register the memory area.

[0116] In this step, the application's memory area is registered with the RDMA device so that the device can directly access these memory areas for data transfer. This is done by using the API or library functions provided by the manufacturer.

[0117] Through the above steps S1041 to S1044, RDMA-related resources can be initialized based on the acquired RDMA configuration, and a reliable RDMA connection can be established.

[0118] Step S105: Synchronize RDMA resource information and interconnection information.

[0119] The operation performed in this step is to synchronize the RDMA resource information and the internal interconnection topology information with the communication peer. By performing this operation, it can be ensured that the RDMA communication parameters are consistent with the internal connection structure between the devices, so that efficient communication can be achieved.

[0120] Synchronization of RDMA resource information and interconnection information refers to ensuring that the devices participating in RDMA communication have consistent configuration and status information in order to achieve efficient and reliable data transmission. This process involves using previously configured RDMA communication channels and registered memory areas to establish an RDMA connection. In the process of establishing an RDMA connection, a handshake and authentication process will be performed between devices to ensure the security and reliability of the connection. The handshake process includes exchanging necessary information. In a specific example of the present application, the information exchanged during the handshake process includes: RDMA network interface IP address, heterogeneous intelligent computing chip virtual address, heterogeneous intelligent computing chip memory address Key, RDMA network interface QP number, and message starting sequence number, etc. This information is used to establish and maintain connections and ensure the correct transmission of data.

[0121] Among them, the RDMA network interface IP address is used to uniquely identify the device in the network to ensure that the data packet can be correctly sent to the target device. The heterogeneous intelligent computing chip virtual address is used to locate the data in the device memory, so that the data can be transferred directly from the source memory to the target memory without the intervention of the CPU. The heterogeneous intelligent computing chip address key is used to verify the security of the connection and ensure that only authorized devices can access specific memory areas. The address key can prevent unauthorized access and data leakage. The RDMA network interface QP number is used to identify the queue pair (Queue Pair) in the RDMA communication channel to ensure that the data can be correctly sent to the target queue. The message start sequence number is used to ensure the orderliness and reliability of the data. Each data packet has a unique sequence number, and the receiver can verify the integrity and order of the data based on the sequence number.

[0122] It is crucial to synchronize this information between the communicating parties because it directly affects the correctness and efficiency of RDMA communication. By synchronizing this information, this application can ensure that each device uses consistent configuration and status during the communication process, thereby avoiding data transmission errors and performance degradation.

[0123] By executing the above steps S101 to S105, the embodiment of the present application can ensure that all necessary components and resources are correctly identified, configured and synchronized before the RDMA communication starts. This can avoid potential communication errors and optimize communication performance. In this step, steps S101 to S105 are usually executed sequentially, but there may be other feasible steps between the steps, which are not limited here.

[0124] Figure 7 FIG. 2 shows an exemplary detailed process of step S200, ie, the RDMA communication data transmission step, in a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention. Figure 7As shown, step S200 may include steps 2.1 to 2.4, which are specifically as follows:

[0125] Figure 7 The RDMA communication transmission process is shown, which includes two parts: the sending end and the receiving end.

[0126] Sending side: Figure 7 In the figure, the sender is the entity on the left, indicating the party that has completed the communication initialization. The sender sends data to the receiver and receives a confirmation signal from the receiver.

[0127] Receiving end: Figure 7 In the example, the receiver is the entity on the right, which indicates the party that has also completed the communication initialization. The receiver receives the data sent by the sender and sends a confirmation signal to the sender.

[0128] The RDMA communication transmission process is as follows:

[0129] Communication initialization is completed: At the beginning of the process, both parties have completed the communication initialization work, including establishing RDMA connection, configuring communication parameters, etc.

[0130] Step 2.1: Receive the PrepareToRecv signal. The receiving end sends a PrepareToRecv signal to the sending direction, indicating that the receiving end is ready to receive data.

[0131] Step 2.2: Send data based on RDMA WRITE. After receiving the PrepareToRecv signal, the sender uses the RDMAWRITE operation to transfer the data directly from the sender's memory to the sender's memory.

[0132] Step 2.3: Send a SendDone signal. After the data is sent, the sender sends a SendDone signal to the receiver, indicating that the data transmission has been completed.

[0133] Step 2.4: Receive the SendDoneAck signal. After receiving the SendDone signal, the sender sends a SendDoneAck signal to the sender, indicating that the receiver has received the data and confirms that the data transmission is complete.

[0134] Data transfer completed: At the end of the process, both communicating parties confirm the completion of the data transfer.

[0135] The specific implementation forms of the communication control signaling in the embodiments of the present application will be introduced below. In the present application, the control signaling involved in the RDMA communication data transmission process includes the following signaling:

[0136] PrepareToRecv signal: The sender sends the PrepareToRecv signal to the receiver, indicating that the sender is ready to receive data.

[0137] SendDone signal: After data transmission is completed, the receiving end sends a SendDone signal to the sending end, indicating that data transmission has been completed.

[0138] SendDoneAck signal: After receiving the SendDone signal, the sender sends a SendDoneAck signal to the receiver, indicating that the sender has received the data and confirms that the data transmission is complete.

[0139] The control signaling designed in this application supports multiple implementation forms, including TCP and RDMA. In view of the poor performance of TCP, this application gives priority to using RDMA WRITE to implement control signaling in some specific implementation methods. A feasible specific implementation method is: the receiving end or the sending end declares a piece of memory specifically for receiving or sending signaling, and each signaling only requires 4 bytes of memory space in the heterogeneous intelligent computing chip, so only 12 bytes of memory space are required to complete a transmission, and it can be reused. The address-related information of each signaling is synchronized between the sending end and the receiving end at the communication initialization node. For example, the address of the PrepareToRecv signal needs to be synchronized from the sending end to the receiving end. During the data transmission process, only one RDMA WRITE operation is required to complete the signaling transmission. At the same time, the other end will poll the value of the corresponding address. When the value of the semaphore is found to be 1, it means that the corresponding signal has been received, so it will proceed to the next step.

[0140] In a specific implementation, the present application sends data based on RDMA WRITE. After the sender receives the PrepareToRecv signal, it uses the RDMA WRITE operation to transfer the data directly from the sender's memory to the sender's memory. RDMA WRITE is a unilateral operation, and the receiver does not need to perform any additional operations to receive data. In other specific implementations, the present application may also use other interfaces, such as READ, SEND / RECV, etc. to send data.

[0141] After completing the RDMA communication transmission, this application will reset the PrepareToRecv signal, SendDone signal, and SendDoneAck signal. This means that after each data transmission is completed, these signals will be reset to the initial state to prepare for the next data transmission. The details are as follows:

[0142] For the PrepareToRecv signal: After each data transmission is completed, the sender resets the PrepareToRecv signal memory space to indicate that it is ready to receive the next data.

[0143] For the SendDone signal: After each data transmission is completed, the receiving end resets the SendDone signal memory space to indicate that it has completed the data transmission.

[0144] For the SendDoneAck signal: After each data transmission is completed, the sender will reset the SendDoneAck signal memory space to indicate that it has received the confirmation signal sent by the receiver.

[0145] Figure 8 The following is an exemplary detailed process of step S300, i.e., the RDMA communication exit step, in a heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention. When the program completes step S200, i.e., completes all data transmission tasks, the RDMA communication exit operation will be executed. The RDMA communication exit process is started to ensure that system resources are correctly released and the system returns to the initial state. Figure 8 As shown, step S300 may include the following key steps:

[0146] Step S301: Close the connection.

[0147] First, the program closes all connections related to RDMA communication. This includes closing the physical connection between the RDMA device and the logical connection established during the communication process, such as Queue Pairs (QPs) or CompletionQueues (CQs).

[0148] Step S302: Release memory.

[0149] Next, the program releases the memory resources used during the RDMA communication process. This includes releasing the buffers used to store data and releasing the memory used to store RDMA communication state information.

[0150] Step S303: Cancel registration.

[0151] Memory registration is used during RDMA communication, and the program also needs to cancel the registration of these heterogeneous intelligent computing chip memory areas. This will make these memory areas no longer associated with the RDMA device and can be used by other programs or system components.

[0152] Step S304: Cleaning status.

[0153] Finally, the program cleans up all state information related to the RDMA communication. This includes clearing any outstanding transfer requests, error status, or other state information related to the communication.

[0154] In the present application, step S400, ie, the RDMA communication exception handling process, is uniquely designed to ensure that the system can correctly handle the exception and take appropriate measures to restore communication or minimize losses.

[0155] Exemplarily, in step S100, i.e., the RDMA communication initialization process of the heterogeneous intelligent computing chip, the present application will detect and process the following initialization anomalies.

[0156] RDMA network connectivity anomaly: During the initialization process, the system needs to ensure the connectivity of the RDMA network. If there is a problem with the network connection, such as link failure, network congestion, etc., communication will fail. The system can detect network connectivity anomalies by sending network detection packets, checking network status, etc.

[0157] Device discovery anomalies: During the initialization process, the system needs to discover and identify the heterogeneous intelligent computing chip devices involved in the communication. If there is a problem in the device discovery process, such as the device not responding or the device configuration is incorrect, communication will not be established. The system can detect device discovery anomalies by sending device discovery requests, checking device status, etc.

[0158] Configuration error: During the initialization process, the system needs to correctly configure the RDMA communication parameters, such as IP address, port number, queue depth, etc. If the configuration is wrong, communication will fail. The system can detect configuration errors by verifying configuration parameters and checking configuration files.

[0159] Insufficient resources: During the initialization process, the system needs to allocate sufficient resources to support RDMA communication, such as memory, CPU, etc. If the resources are insufficient, communication performance will be degraded or communication will fail. The system can detect insufficient resources anomalies by monitoring resource usage and checking resource allocation policies.

[0160] In another specific implementation scheme, during the RDMA communication data transmission process of the heterogeneous intelligent computing chip in step S200, the embodiment of the present application will detect and process the following data transmission anomalies.

[0161] Abnormal packet loss: During the transmission process, if the data packet is lost or damaged, it will cause communication errors or data loss. This application detects abnormal packet loss by monitoring the transmission status of the data packet and checking the receiving status of the receiving end.

[0162] Link interruption: During the transmission process, if the network connection is interrupted or unstable, it will cause communication interruption or increased delay. This application detects link interruption by monitoring the status of the network connection and checking the operating status of the network equipment.

[0163] Congestion control: During the transmission process, if the network is congested or overloaded, it will cause increased communication delay or communication failure. The embodiment of the present application detects and handles congestion control anomalies by monitoring network traffic, adjusting transmission parameters, etc.

[0164] Data consistency: During the transmission process, if the consistency of data is damaged, it will lead to calculation errors or data loss. The embodiment of the present application detects and handles data consistency anomalies by implementing data verification, data redundancy, etc.

[0165] Furthermore, in step S300, during the heterogeneous intelligent computing chip RDMA communication exit phase, the embodiment of the present application detects and processes the following anomalies.

[0166] Abnormal resource release: During the exit phase, the system needs to release resources used during the communication process, such as memory, CPU, etc. If there are problems in the resource release process, such as resource leakage, resource competition, etc., it will lead to system resource waste or system instability.

[0167] Connection closing exception: During the exit phase, the system needs to close all connections related to RDMA communication. If there is a problem in the connection closing process, such as the connection is not closed, the connection status is inconsistent, etc., it will lead to waste of system resources or system instability. The system can detect connection closing exceptions by monitoring the connection status and checking the connection closing policy.

[0168] Configuration recovery anomaly: During the exit phase, the system needs to be restored to the initial configuration state. If there is a problem in the configuration recovery process, such as the configuration is not restored or the configuration is wrong, the system will not be able to operate normally. The system can detect configuration recovery anomalies by verifying configuration parameters and checking configuration recovery policies.

[0169] Fig. 9 A schematic diagram of a hybrid communication SDK software system running on an intelligent computing server according to another embodiment of the present invention is shown. The SDK software system mainly includes the following modules:

[0170] Heterogeneous intelligent computing chip RDMA communication SDK is used to provide users with programming interfaces for communication operations. This is the top-level module of the entire system. Among them, the programming interface is used to send and receive data. It includes synchronous interfaces (Send / Recv) and asynchronous interfaces (AsyncSend / AsyncRecv).

[0171] Communication initialization module: responsible for initializing the communication environment, including establishing communication channels, configuring communication parameters, etc.

[0172] The data transmission module is used for data transmission, mainly including sending data and receiving data.

[0173] The communication exit module is responsible for releasing communication resources after the communication is completed, including closing the communication channel, releasing memory, etc.

[0174] The communication exception handling module is used to handle abnormal situations that may occur during the communication process, such as communication errors, timeouts, etc.

[0175] The entire system of the embodiment of the present application achieves efficient communication between heterogeneous intelligent computing chips through the collaborative work of these modules, providing users with a convenient programming interface and stable communication services.

[0176] Another embodiment of the present invention provides a heterogeneous intelligent computing chip RDMA communication system, the system comprising:

[0177] Heterogeneous intelligent computing chip RDMA communication initialization unit, used to perform RDMA communication initialization to identify intelligent computing chips and discover the internal interconnection topology;

[0178] A heterogeneous intelligent computing chip RDMA communication data transmission unit is used to perform RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology;

[0179] A heterogeneous intelligent computing chip RDMA communication exit operation unit, used to perform an RDMA communication exit operation to release the occupied intelligent computing chip resources;

[0180] The heterogeneous intelligent computing chip RDMA communication anomaly detection and processing unit is used to perform RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and to process the identified communication anomaly event.

[0181] The heterogeneous intelligent computing chip RDMA communication system of the embodiment of the present invention can be used to execute the RDMA communication method described in the above embodiment.

[0182] According to an exemplary embodiment of the present invention, a heterogeneous intelligent computing chip platform is also provided. The heterogeneous chip hybrid platform can support heterogeneous chips from different manufacturers such as AMD, Huawei Ascend, Tianshu Zhixin, Muxi, Moore Thread, NVIDIA, etc., and can be configured with a desired number (e.g., 1,000 or 10,000 cards) of heterogeneous chips. The data communication in the heterogeneous intelligent computing chip platform adopts the heterogeneous chip RDMA communication method described in the above embodiment.

[0183] In one embodiment, the communication in the heterogeneous intelligent computing chip platform performs the following steps:

[0184] Step S100: Execute RDMA communication initialization to identify the intelligent computing chip and discover the internal interconnection topology;

[0185] Step S200: performing RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology;

[0186] Step S300: Execute the RDMA communication exit operation to release the occupied intelligent computing chip resources;

[0187] Step S400: performing RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and process the identified communication anomaly event.

[0188] In addition, the communication in the heterogeneous intelligent computing chip platform also executes the other steps of the above-mentioned heterogeneous intelligent computing chip RDMA communication method, which will not be repeated here.

[0189] In addition, according to another embodiment of the present invention, a computer-readable storage medium is also provided, on which program instructions are stored, and when the program instructions are run by a computer or a processor, the corresponding steps of the heterogeneous intelligent computing chip RDMA communication method of the embodiment of the present invention are used to execute. The storage medium may include, for example, a memory card of a smart phone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disk read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0190] In one embodiment, the computer program instructions, when executed by a computer, can execute the heterogeneous intelligent computing chip RDMA communication method according to an embodiment of the present invention.

[0191] In one embodiment, the computer program instructions, when executed by a computer, perform the following steps:

[0192] Step S100: Execute RDMA communication initialization to identify the intelligent computing chip and discover the internal interconnection topology;

[0193] Step S200: performing RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology;

[0194] Step S300: Execute the RDMA communication exit operation to release the occupied intelligent computing chip resources;

[0195] Step S400: performing RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and process the identified communication anomaly event.

[0196] In addition, when the computer program instructions are executed by the computer, the other steps of the above-mentioned heterogeneous intelligent computing chip RDMA communication method are also executed, which will not be repeated here.

[0197] According to the heterogeneous intelligent computing chip RDMA communication method, system, platform and storage medium of the embodiments of the present invention, by adopting the RDMA communication method for hybrid communication of heterogeneous intelligent computing chips, communication performance can be guaranteed, thereby improving computing resource utilization.

[0198] The technology of this application has the following advantages over the existing technology: First, it has a wider range of application scenarios, is not limited to the same manufacturer or software ecosystem, and can cope with more complex heterogeneous chip scenarios. Second, it provides higher performance and can provide better computing resource utilization by using a high-performance RDMA communication protocol solution.

[0199] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present invention thereto. Various changes and modifications may be made therein by one of ordinary skill in the art without departing from the scope and spirit of the present invention. All such changes and modifications are intended to be included within the scope of the present invention as required by the appended claims.

[0200] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0201] The various techniques described herein may be implemented in combination with hardware or software, or a combination thereof. Thus, the method, apparatus, and device of the present invention, or some aspects or portions of the method, apparatus, and device of the present invention may be in the form of program codes (i.e., instructions) embedded in a tangible medium, such as a removable hard disk, a USB flash drive, a floppy disk, a CD-ROM, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes a device for practicing the present invention.

[0202] In the case where the program code is executed on a programmable computer, the electronic device generally includes a processor, a storage medium readable by the processor (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device. The memory is configured to store the program code; the processor is configured to execute the remote assistance method of the present invention according to the instructions in the program code stored in the memory.

[0203] By way of example and not limitation, readable media include readable storage media and communication media. Readable storage media stores information such as computer-readable instructions, data structures, program modules or other data. Communication media generally embody computer-readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and include any information transmission medium. Any combination of the above is also included in the scope of readable media. In the specification of the present application, algorithms and displays are not inherently related to any specific computer, virtual system or other device. Various general systems can also be used with the examples of the present invention. According to the above description, it is obvious to construct the structure required for such systems. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be used to implement the software and algorithm-related content of the present invention described herein, and the description of the specific language above is only to disclose some specific embodiments of the present invention, but not to limit the use of the specific language.

[0204] In the specification of the application, a lot of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this specification.

[0205] In the several embodiments provided in this application, it should be understood that the disclosed devices, equipment and methods can be implemented in other ways. For example, the above-described device and equipment embodiments are only illustrative, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0206] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0207] Similarly, it should be understood that in order to streamline the present invention and help understand one or more of the various inventive aspects, in the description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present invention should not be interpreted as reflecting the following intention: the claimed invention requires more features than the features explicitly stated in each claim. More specifically, as reflected in the corresponding claims, the inventive point is that the corresponding technical problem can be solved with less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present invention.

[0208] It will be understood by those skilled in the art that, except for mutually exclusive features, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this specification may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0209] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims, any one of the claimed embodiments may be used in any combination.

[0210] The components in the various embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that computers, servers, microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules in the equipment and devices according to embodiments of the present invention. The present invention can also be implemented as a device program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0211] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. Nouns such as "comprise", "include", "have" do not exclude the presence of elements or steps not listed in the claims. The quantifier "one" or "a" preceding an element does not exclude the presence of multiple such elements. The present invention can be implemented by means of hardware including several different elements and by means of appropriately programmed computers. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order, and these words may be interpreted as names.

[0212] In addition, in the embodiments of the present application, the expression "and / or" includes any and all combinations of the associated listed words. For example, the expression "A and / or B" may include A, may include B, or may include both A and B. Words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "multiple" refers to two or more.

[0213] In the description of the embodiments of the present application, when a component is referred to as being "connected" or "accessed" to other components, it should be understood that: the component is not only directly connected to or accessed to other components, but also another component may exist between the component and the other components. On the other hand, when a component is referred to as being "directly connected" or "directly accessed" to other components, it should be understood that there is no component between them. When a component is referred to as being "movably connected" to other components, it means that the positional relationship between the component and the other components is changeable, such as the relationship between a slide rail and a slider, the slider can slide on the slide rail, and the slide rail and the track are movably connected. When a component is referred to as being "fixedly connected" to other components, it means that the positional relationship between the component and the other components after being assembled together is relatively fixed.

[0214] The above is only a specific embodiment of the present invention or an explanation of a specific embodiment. The protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Although the present application is described according to a limited number of embodiments, it is understood by the technicians in the technical field that other embodiments can be envisioned within the scope of the present application described thereby, thanks to the above description. In addition, it should be noted that the language used in this specification is mainly selected for the purpose of readability and teaching, rather than for explaining or limiting the subject matter of the present application. Therefore, without departing from the scope and spirit of the attached claims, many modifications and changes are obvious to ordinary technicians in the technical field. For the scope of the present application, the disclosure made in the present application is illustrative, not restrictive, and the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A heterogeneous intelligent computing chip RDMA communication method, characterized in that: The method comprises: Step S100: Execute RDMA communication initialization to identify the intelligent computing chip and discover the internal interconnection topology; Step S200: performing RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology; Step S300: Execute the RDMA communication exit operation to release the occupied intelligent computing chip resources; Step S400: performing RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and process the identified communication anomaly event.

2. The method according to claim 1, characterized in that The step S100 includes: Step S101: performing an intelligent computing chip discovery operation to identify and discover available intelligent computing chips; Step S102: performing an intra-machine interconnection topology discovery operation to determine the interconnection topology relationship between the discovered intelligent computing chips; Step S103: performing a network interface discovery operation to identify a network interface that can be used for RDMA communication; Step S104: Execute RDMA resource initialization operation; Step S105: performing synchronization operations on the RDMA resource information and the interconnection information.

3. The method according to claim 2, characterized in that The step S101 is implemented by polling PCIe, and includes the following steps: Step S1011: Initialize the PCIe polling mechanism to periodically read PCIe device information; Step S1012: performing a PCIe polling operation, and reading PCIe device information in each polling cycle, wherein the PCIe device information includes a device list and configuration information; Step S1013: Analyze the read PCIe device information to determine whether there is an intelligent computing chip.

4. The method according to claim 2, characterized in that The step S102 includes: Step S1021: performing a connection relationship detection operation to determine the physical connection mode between the intelligent computing chips in the server; Step S1022: constructing an interconnection topology structure inside the server based on the physical connection method.

5. The method according to claim 2, characterized in that The step S103 includes: Step S1031: performing a protocol identification operation to identify a network protocol supporting RDMA in the system; Step S1032: searching for devices supporting RDMA in the network according to the identified network protocol; Step S1033: Acquire the configuration information of the device supporting RDMA, wherein the configuration information includes an IP address and a port number.

6. The method according to claim 2, characterized in that The step S104 includes: Step S1041: performing RDMA initialization on the intelligent computing chip to ensure that it works normally and supports RDMA communication; Step S1042: creating an RDMA communication channel for the intelligent computing chip that has completed RDMA initialization; Step S1043: performing parameter configuration on the created RDMA communication channel, wherein the configuration includes the peer QP number and the peer IP address; Step S1044: Execute a memory area registration operation to register the memory area of ​​the application program in the intelligent computing chip, so that the intelligent computing chip can directly access the memory area for data transmission.

7. The method according to claim 1, characterized in that The step S200 includes: Receive and identify the PrepareToRecv signal from the intelligent computing chip as the receiving end; Sending the data to be processed to the intelligent computing chip based on the RDMA WRITE operation, so as to store the data to be processed into the memory of the receiving end; After the data to be processed is sent, a SendDone signal is sent to the intelligent computing chip; Receive and identify the SendDoneAck signal from the intelligent computing chip to end data transmission.

8. The method according to claim 7, characterized in that In the step S200, a memory dedicated to receiving signaling is set for the receiving end.

9. The method according to claim 1, characterized in that The step S300 includes: Step S301: closing the physical connection and logical connection related to the RDMA communication; Step S302: releasing memory resources used in the RDMA communication process; Step S303: canceling the registration of the memory area of ​​the intelligent computing chip during the RDMA communication process; Step S304: clearing the status information related to the RDMA communication.

10. The method according to any one of claims 1 to 9, characterized in that: The step S400 includes: In the step S100, initialization anomaly detection and processing is performed, wherein the initialization anomaly includes at least one of RDMA network connectivity anomaly, device discovery anomaly, configuration error, and insufficient resources; and / or, In the step S200, data transmission anomaly detection and processing are performed, wherein the data transmission anomaly includes at least one of abnormal packet loss, link interruption, congestion control, and data consistency; and / or, In the step S300, communication exit anomaly detection and processing is performed, wherein the communication exit anomaly includes at least one of resource release anomaly, connection closing anomaly, and configuration recovery anomaly.

11. A heterogeneous intelligent computing chip RDMA communication system, characterized in that: The system comprises: Heterogeneous intelligent computing chip RDMA communication initialization unit, used to perform RDMA communication initialization to identify intelligent computing chips and discover the internal interconnection topology; A heterogeneous intelligent computing chip RDMA communication data transmission unit is used to perform RDMA communication data transmission on the data to be processed based on the identified intelligent computing chip and the discovered in-machine interconnection topology; A heterogeneous intelligent computing chip RDMA communication exit operation unit, used to perform an RDMA communication exit operation to release the occupied intelligent computing chip resources; The heterogeneous intelligent computing chip RDMA communication anomaly detection and processing unit is used to perform RDMA communication anomaly detection and processing operations to identify whether a communication anomaly event occurs and to process the identified communication anomaly event.

12. A heterogeneous intelligent computing chip computing platform, characterized in that: It comprises at least two heterogeneous intelligent computing chips, wherein data communication in the heterogeneous intelligent computing chip platform adopts the method as described in any one of claims 1 to 10.

13. A readable storage medium storing program instructions, when the program instructions are read and executed by a computing device, the computing device executes the method according to any one of claims 1 to 10.

Citation Information

Patent Citations

  • Method, system and device for optimizing deep learning heterogeneous resource communication efficiency and medium

    CN114281521A

  • Systems, apparatuses, and methods for streaming input / output data

    CN114490499A

  • RDMA bandwidth transmission test method and system and storage medium

    CN115037651A

  • Remote split infrastructure processing unit (IPUS)

    CN115858442A

  • Data processing method and system, computing equipment and storage medium

    CN116402673A