Token computing power network forwarding enhancement method and device of intelligent computing cloud platform

Abstract

The application provides a Token computing power network forwarding enhancement method and device of an intelligent computing cloud platform, relates to the technical field of intelligent computing centers, intelligent computing centers and computing power infrastructure, and the Token computing power network forwarding enhancement method of the intelligent computing cloud platform comprises the following steps: S1, determining a GPU node group for executing a task from a Token computing power network according to a task instruction, wherein the GPU node group comprises a plurality of target GPU nodes; S2, generating a group identifier and a group forwarding table of the GPU node group; S3, for each target GPU node, in response to the target GPU node having to-be-distributed data, publishing the group identifier and the to-be-distributed data in the Token computing power network; and S4, calling a forwarding node of the Token computing power network and forwarding the to-be-distributed data according to the group identifier and the group forwarding table. The application can improve the flexibility of the computing power platform, improve the expansion efficiency of the computing power center cluster, improve the transmission efficiency of the Token computing power network and the speed of Token computing, and reduce the delay of Token output.

Description

Technical Field

[0001] This invention relates to the fields of intelligent computing centers, smart computing centers, and computing infrastructure technology, specifically to a method and apparatus for enhancing token computing power network forwarding in an intelligent computing cloud platform. Background Technology

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "smart computing centers" have emerged.

[0003] An "intelligent computing center" refers to a facility that provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models) by utilizing large-scale heterogeneous computing resources, including general-purpose and intelligent computing power. Intelligent computing centers encompass facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] "Intelligent computing center" includes, but is not limited to, "intelligent computing center".

[0005] "Intelligent computing center" or artificial intelligence computing center is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing center" and "smart computing center". It is the ability of computer equipment or computing / data center to process parameters. It is the ability of computer hardware and software to work together to execute a certain computing requirement. It is the computing power to achieve the target result output by processing parameter data. It is a new type of productivity that integrates parameter computing power, network carrying capacity and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] In existing technologies, network segment planning is usually a fixed setting. If business growth or a surge in computing power demand necessitates scaling up, the network configuration needs to be modified node by node. Computing resources communicate through paths in the computing network. During network forwarding, if the data forwarding path between the source and destination is circuitous, it may increase hop-by-hop processing latency and bandwidth consumption. Furthermore, repeatedly storing or transmitting the same data blocks will lead to file redundancy overhead, potentially reducing the transmission efficiency of the token computing network, slowing down token computation, causing token output latency, and leading to network congestion. This results in insufficient resilience of the computing platform when facing sudden computing power demands, and reduces the efficiency of cluster scaling in the computing center.

[0008] It is evident that since the emergence of intelligent computing centers, how to improve the elasticity of computing power platforms, increase the efficiency of computing power center cluster expansion, optimize network forwarding paths, improve the transmission efficiency of token computing networks and the speed of token calculation, reduce the latency of token output, and reduce network congestion has always been a pressing problem to be solved in this field. Summary of the Invention

[0009] This invention provides a method and apparatus for enhancing the forwarding of token computing power networks in an intelligent computing cloud platform, in order to solve the problems in the prior art that reduce the transmission efficiency of token computing power networks, reduce the speed of token computing, cause delays in token output and network congestion, make the computing power platform insufficiently elastic when facing sudden computing power demands, and reduce the efficiency of cluster expansion of computing power centers.

[0010] To solve the above problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for enhancing the Token computing power network forwarding of an intelligent computing cloud platform, comprising: Step S1: According to the task instructions, determine the GPU node group to execute the task from the Token computing power network. The GPU node group includes multiple target GPU nodes. Step S2: Generate the group identifier and group forwarding table for the GPU node group; Step S3: For each target GPU node, in response to the target GPU node having data to be distributed, publish the group identifier and the data to be distributed in the Token computing power network; Step S4: Call the forwarding node of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

[0011] In one embodiment, step S2 includes: Step S2.1: Generate the group identifier of the GPU node group and establish the association between each target GPU node and the group identifier; Step S2.2: Call each target GPU node to publish group identifier indication information on the Token computing power network. The group identifier indication information is used to indicate the group identifier associated with the target GPU node. Step S2.3: Call the forwarding node in the computing power network to receive the group identifier indication information, and obtain the group forwarding table according to the forwarding path of the group identifier information.

[0012] In one embodiment, step S2.3 includes: Step S2.3.1: Obtain the source information and interface information of the group identifier; Step S2.3.2: Obtain the group forwarding table based on the forwarding path of the group identifier information, the source information of the group identifier, and the interface information.

[0013] In one embodiment, the forwarding node includes leaf nodes and spine nodes, and step S4 includes: Step S4.1: In response to the leaf node receiving the data to be distributed and the group identifier, the leaf node looks up the group forwarding table based on the group identifier and obtains the first forwarding path; Step S4.2: The leaf node sends the data to be distributed and the group identifier to the GPU node or ridge node pointed to by the first forwarding path.

[0014] In one embodiment, step S4.2 includes: Step S4.2.1: In response to the first forwarding path including multiple forwarding paths, the data to be distributed and the group identifier are copied according to the number of the first forwarding paths; Step S4.2.2: Send the copied data to be distributed and the group identifier to the GPU node or spine node pointed to by each first forwarding path.

[0015] In one embodiment, it also includes: Step S4.3: In response to the spine node receiving the data to be distributed and the group identifier, the spine node looks up the group forwarding table based on the group identifier and obtains the second forwarding path; Step S4.4: The spine node forwards the data to be distributed and the group identifier to the leaf node pointed to by the second forwarding path.

[0016] In one embodiment, step S4.4 includes: Step S4.4.1: In response to the fact that the second forwarding path includes multiple forwarding paths, perform a hash function calculation for each second forwarding path; Step S4.4.2: Determine the target forwarding path from the second forwarding paths based on the hash function calculation results of each second forwarding path; Step S4.4.3: Forward the data to be distributed and the group identifier to the leaf node pointed to by the target forwarding path.

[0017] In one embodiment, it also includes: Step S5: In response to the release of any target GPU node, call the forwarding node to update the group forwarding table to delete the forwarding path corresponding to the released target GPU node.

[0018] In one embodiment, it also includes: Step S6: In response to any candidate GPU node joining the GPU node group, establish the association between the candidate GPU node and the group identifier; Step S7: Call the forwarding node to update the group forwarding table to add forwarding paths corresponding to candidate GPU nodes.

[0019] Secondly, the present invention also provides a token computing power network forwarding enhancement device for an intelligent computing cloud platform, comprising: The node determination module is used to determine the GPU node group to execute the task from the Token computing power network according to the task instructions. The GPU node group includes multiple target GPU nodes. The information generation module is used to generate group identifiers and group forwarding tables for GPU node groups; The data publishing module is used to publish the group identifier and the data to be distributed in the Token computing power network for each target GPU node in response to the target GPU node having data to be distributed. The data forwarding module is used to call the forwarding nodes of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

[0020] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps in the Token computing power network forwarding enhancement method of the intelligent computing cloud platform as described in the first aspect above.

[0021] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the Token computing power network forwarding enhancement method of the intelligent computing cloud platform as described in the first aspect above.

[0022] Fifthly, the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the Token computing power network forwarding enhancement method of the intelligent computing cloud platform as described in the first aspect above.

[0023] In this invention, step S1 involves determining the GPU node group to execute the task from the Token computing power network according to the task instruction. The GPU node group includes multiple target GPU nodes. Step S2 involves generating a group identifier and a group forwarding table for the GPU node group. Step S3 involves publishing the group identifier and the data to be distributed in the Token computing power network for each target GPU node, in response to the target GPU node having data to be distributed. Step S4 involves calling the forwarding nodes of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table. This invention can improve the elasticity of the computing power platform, improve the efficiency of the computing power center cluster expansion, optimize the network forwarding path, improve the transmission efficiency and token calculation speed of the Token computing power network, reduce the latency of token output, and reduce network congestion. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a method for enhancing the Token computing power network forwarding of an intelligent computing cloud platform provided by the present invention; Figure 2 This is a flowchart of a method for generating a group identifier and a group forwarding table for GPU node groups, provided by the present invention; Figure 3 This invention provides a flowchart for forwarding data to be distributed based on a group identifier and a group forwarding table; Figure 4 This invention provides a flowchart for forwarding data to be distributed; Figure 5 This invention provides a flowchart for updating a group forwarding table; Figure 6 This is a structural diagram of a token computing power network forwarding enhancement device for an intelligent computing cloud platform provided by the present invention; Figure 7 This is a structural diagram of an electronic device provided by the present invention. Detailed Implementation

[0026] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The “computing power” mentioned in this invention refers to: the ability of computer equipment or computing / data center to process information; the ability of computer hardware and software to work together to perform a certain computing requirement; the computing power to achieve the target result output by processing information data; and a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0028] The "computational power" (CP) described in this invention refers to the ability of a data center server to process data and output results. It is a comprehensive indicator of a data center's computing power, encompassing general computing power, supercomputing power, and intelligent computing power. The commonly used unit of measurement is floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), with higher values ​​indicating stronger overall computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A supercomputers, 500,000 mainstream server CPUs, or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0029] The "Network Power" (NP) mentioned in this invention refers to the performance of data transmission capability of computing facilities, which includes comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, and involves network transmission within and between data centers. It is a comprehensive indicator for measuring network transmission scheduling capability.

[0030] The "Storage Power" (SP) described in this invention refers to the comprehensive capabilities of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon operation. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and internal storage devices within servers. The commonly used unit of measurement for storage capacity is exabytes (EB, 1EB = 2^60 bytes), while the commonly used unit of measurement for performance is the number of read / write operations per second (IOPS / TB). Disaster recovery ratio is an important indicator of security and reliability.

[0031] The "computing infrastructure" mentioned in this invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, enabling centralized computing, storage, transmission, and application of information.

[0032] The "new information infrastructure" mentioned in this invention refers to network infrastructure such as 5G networks, fiber optic broadband networks, backbone networks, international communication networks, and satellite internet; computing infrastructure such as data centers, general computing centers, intelligent computing centers, and supercomputing centers; and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0033] The “computing power” mentioned in this invention includes: general computing power, intelligent computing power, and supercomputing power.

[0034] The "general computing power" mentioned in this invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0035] The "intelligent computing power" mentioned in this invention refers to: a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for various artificial intelligence innovative applications, such as natural language processing and machine vision.

[0036] The “supercomputing power” mentioned in this invention refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0037] The "intelligent computing center" described in this invention refers to a facility that, through the use of large-scale heterogeneous computing resources, including general-purpose computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), primarily provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models). The intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0038] The "intelligent computing cloud platform" mentioned in this invention, abbreviated as "intelligent computing cloud", refers to a cloud computing platform that integrates hardware and software resources based on an intelligent computing center.

[0039] The "intelligent computing center" mentioned in this invention includes, but is not limited to, "smart computing center".

[0040] The "intelligent computing center" mentioned in this invention, also known as an artificial intelligence computing center, is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0041] The "computing center" mentioned in this invention refers to a facility that is mainly composed of infrastructure such as wind, thermal, hydro, and electricity, and IT hardware and software equipment, and has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0042] The "supercomputing center" mentioned in this invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters. It can provide large-scale computing, storage and network services and is widely used in aerospace, defense, oil exploration, climate modeling and genome sequencing and other application scenarios.

[0043] The “computing resources” mentioned in this invention refer to the technologies and facilities required for the development of the digital society that have the ability to compute, transmit, store and apply information, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guaranteeing resources such as wind, fire, water and electricity.

[0044] The “computing power network” mentioned in this invention refers to the platform network on which multiple GPUs exchange a large amount of data during the token calculation process (training or inference).

[0045] The “Token” mentioned in this invention refers to the basic unit of measurement when a large language model or multimodal large model processes text, code, image descriptions, document content, or other serialized inputs and outputs.

[0046] It is important to emphasize that different models use different tokenizers. The same text may correspond to different numbers of tokens in different models.

[0047] Please see Figure 1 , Figure 1 This is a flowchart of a method for enhancing the Token computing power network forwarding of an intelligent computing cloud platform provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step S1: According to the task instructions, determine the GPU node group to execute the task from the Token computing power network. The GPU node group includes multiple target GPU nodes.

[0048] The token factory hosted on the intelligent computing cloud platform involves a large number of data distribution operations during token production, whether in training or inference, such as global reduction (allreduce), full collection (allgather), and broadcast. These operations require sending data from one GPU to multiple GPUs.

[0049] Allreduce, Allgather, and Broadcast are all collection communication operations in distributed computing, used for data synchronization between multiple nodes / cards.

[0050] In one embodiment, based on task instructions, the computing power reserve, memory bandwidth, communication latency, and current load status of all available GPU nodes in the Token computing power network are evaluated in real time. The task requirements are dynamically matched through an intelligent scheduling algorithm, thereby selecting a group of collaborative and efficient GPU nodes.

[0051] In one embodiment, multiple GPU nodes participating in the computation can be determined by task instructions and a task scheduler, forming a group of GPU nodes that execute the task.

[0052] The technical effect of step S1 is that it can accurately select a group of GPU nodes with better performance and higher efficiency, ensuring that the computing resources of the Token computing power network are allocated efficiently and reasonably.

[0053] Step S2: Generate the group identifier and group forwarding table for the GPU node group.

[0054] In one embodiment, a globally unique group ID is generated for the GPU node group to clearly distinguish different computing groups in the network. That is, all GPU nodes participating in this computation negotiate a group ID and publish the group ID they have joined in the Token computing power network. The forwarding nodes in the Token computing power network build and maintain the corresponding group forwarding table, which records the forwarding path of the group ID in detail.

[0055] The technical effect of step S2 is to ensure efficient forwarding and accurate addressing of data interaction within the group, enabling data packets to be forwarded quickly within the Token computing network, significantly reducing the latency and overhead of cross-node communication, and improving the transmission efficiency of the Token computing network and the speed of Token calculation.

[0056] Step S3: For each target GPU node, in response to the target GPU node having data to be distributed, publish the group identifier and the data to be distributed in the Token computing power network.

[0057] In one embodiment, for each target GPU node, in response to the target GPU node's local cache or computation output containing intermediate results or parameter data to be distributed, the node packages its own group identifier with the data to be distributed and efficiently publishes it to the Token computing power network through a unified communication protocol.

[0058] The technical effects of step S3 are: shortening the propagation path and distribution latency of cross-node data, reducing the latency of token output, improving the data synchronization efficiency in multi-node collaborative training or inference scenarios, and ensuring the data consistency and task continuity of the token computing power network.

[0059] Step S4: Call the forwarding node of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

[0060] In one embodiment, the forwarding node retrieves the corresponding group forwarding table based on the group identifier, and accurately forwards the data to be distributed to the target node group according to the forwarding path recorded in the table.

[0061] The technical effects of step S4 are: to improve the efficiency of scaling up the computing center cluster, optimize network forwarding paths, improve the transmission efficiency of the token computing network and the speed of token calculation, reduce the latency of token output, reduce network congestion, and improve the elasticity of the computing platform.

[0062] Please see Figure 2 , Figure 2 This is a flowchart of a process for generating a group identifier and a group forwarding table for a GPU node group, as provided by the present invention. Figure 2 As shown, step S2 includes: Step S2.1: Generate the group identifier of the GPU node group and establish the association between each target GPU node and the group identifier.

[0063] In one embodiment, a globally unique group ID is generated for the GPU node group to clearly distinguish different computing groups in the network; that is, all GPU nodes participating in this computation negotiate a group ID.

[0064] Furthermore, a mapping relationship is established between each target GPU node in the group and the group identifier, and this mapping relationship is recorded and stored for quick querying during subsequent routing and data distribution.

[0065] The technical effects of step S2.1 are: avoiding forwarding errors caused by confusing node identities or ambiguous group boundaries in the Token computing power network, improving the accuracy of Token output, increasing response speed, and enabling dynamic addition and deletion of Token computing power network nodes without affecting the overall network topology. This provides a foundation for the elastic scaling of the computing power platform and enhances scalability and ease of operation and maintenance.

[0066] Step S2.2: Call each target GPU node to publish group identifier indication information on the Token computing power network. The group identifier indication information is used to indicate the group identifier associated with the target GPU node.

[0067] The target GPU node publishes the group identifier it has joined to the Token computing power network. Optionally, each target GPU node is invoked to trigger it to broadcast group identifier indication information in the Token computing power network. This indication information is presented in the form of a lightweight heartbeat message or packet, and the group identifier indication information is used to indicate the group identifier (Group ID) currently associated with the target GPU node, so that forwarding nodes can perceive its group affiliation.

[0068] In one embodiment, the target GPU node can proactively broadcast its associated group identifier to the forwarding nodes in the Token computing power network at a set time interval to ensure the continuous synchronization and updating of node information.

[0069] The technical effects of step S2.2 are as follows: It realizes the active announcement of group identifier indication information, enabling each forwarding node of the Token computing power network to know the group identifier that the target GPU node has joined; through the periodic release strategy, the convergence time after the node state of the Token computing power network changes is greatly shortened, which helps to improve the speed of Token calculation, enhances the adaptability of the Token computing power network to node joining, leaving or failure, and lays a real-time and reliable information foundation for the accurate routing of subsequent data distribution.

[0070] Step S2.3: Call the forwarding node in the computing power network to receive the group identifier indication information, and obtain the group forwarding table according to the forwarding path of the group identifier information.

[0071] In one embodiment, step S2.3 includes steps S2.3.1 and S2.3.2, which are further described below.

[0072] Step S2.3.1: Obtain the source information and interface information of the group identifier.

[0073] In one embodiment, after the target GPU node publishes group identifier indication information on the Token computing power network, each forwarding node obtains the group identifier source information and the interface information corresponding to the group identifier through a pre-configured interface and / or protocol.

[0074] Optionally, the group identifier source information may include at least one of the following: the node that generated the group identifier, the publication time, and the version sequence; the interface information may include at least one of the following: the data distribution port, the communication protocol type, and the access control policy. The source information and interface information are used to indicate the access and forwarding capabilities of group-related resources.

[0075] Step S2.3.2: Obtain the group forwarding table based on the forwarding path of the group identifier information, the source information of the group identifier, and the interface information.

[0076] In one embodiment, a forwarding node in the network receives group identifier indication information, records and saves the forwarding path of the group identifier indication information, and the forwarding path of the group identifier indication information is also the forwarding path of the group identifier.

[0077] Forwarding nodes aggregate group identifiers, forwarding paths, source information, and interface information to form a group forwarding table.

[0078] The technical effect of step S2.3 is that by integrating group identifier forwarding path, source information and interface information, the adaptive construction of the group forwarding table is realized, which improves the transmission efficiency of the Token computing power network and provides a reliable data foundation for fine-grained traffic scheduling and fault domain isolation in the Token computing power network, thereby improving the overall communication efficiency and system robustness.

[0079] Please see Figure 3 , Figure 3 This invention provides a flowchart for forwarding data to be distributed based on a group identifier and a group forwarding table, such as... Figure 3 As shown, step S4 includes: Step S4.1: In response to the leaf node receiving the data to be distributed and the group identifier, the leaf node looks up the group forwarding table based on the group identifier and obtains the first forwarding path.

[0080] In one embodiment, the forwarding nodes include leaf nodes and spine nodes. The spine node is the backbone link node, i.e., the spine node; the leaf node is the end node without child nodes, i.e., the leaf node.

[0081] When the GPU has data to distribute to all other nodes currently executing tasks, it only needs to send one copy of the data. When the leaf node receives the data to be distributed and the group identifier, it will copy and distribute the data according to the group forwarding table.

[0082] In one embodiment, when a leaf node in the Token computing power network receives data to be distributed and its associated group identifier sent from a forwarding node or GPU node, it uses the group identifier as an index to perform a matching query in the group forwarding table, quickly retrieves the first forwarding path corresponding to the group identifier, and then initiates data packet encapsulation and forwarding operations based on the path.

[0083] Optionally, the forwarding path includes at least one of the following parameters: next-hop node address, outgoing interface number, priority, and QoS policy.

[0084] The technical effects of step S4.1 are: accurate data distribution is achieved; by using a group forwarding table, the dwell time and processing latency of data in the Token computing power network nodes are significantly shortened, thereby reducing the latency of Token output and improving the transmission efficiency of the Token computing power network.

[0085] Step S4.2: The leaf node sends the data to be distributed and the group identifier to the GPU node or ridge node pointed to by the first forwarding path.

[0086] In one embodiment, step S4.2 includes steps S4.2.1 and S4.2.2, which are further described below.

[0087] Step S4.2.1: In response to the first forwarding path including multiple forwarding paths, the data to be distributed and the group identifier are copied according to the number of the first forwarding paths.

[0088] When the first forwarding path contains multiple optional paths, it indicates that the leaf node is connected to multiple GPU nodes in the Token computing power network that need to receive the data to be distributed. The leaf node determines the number of data copies based on the number of paths in the first forwarding path list, and uses zero-copy or other copying methods to efficiently copy the data to be distributed and the group identifier, ensuring that each copy of the data to be distributed carries the corresponding group identifier.

[0089] Step S4.2.2: Send the copied data to be distributed and the group identifier to the GPU node or spine node pointed to by each first forwarding path.

[0090] The copied data to be distributed and the group identifier are sent to the next-hop node pointed to by each first forwarding path. If the path is direct, the next hop may be the target GPU node within the group; if the path requires multiple layers of switching, the next hop may be the spine node responsible for relaying.

[0091] The technical effect of step S4.2: Leaf nodes independently make search decisions based on group identifiers, enhancing the scalability and fault tolerance of the Token computing power network. Even if the control plane experiences a brief loss of connection, leaf nodes can still continue to complete forwarding tasks based on existing forwarding tables, ensuring the continuous and smooth flow of data in the Token computing power network. Sending operations along multiple paths do not block each other, maximizing the utilization of the Token computing power network's link bandwidth resources and optimizing network forwarding paths.

[0092] Step S4.3: In response to the ridge node receiving the data to be distributed and the group identifier, the ridge node looks up the group forwarding table based on the group identifier and obtains the second forwarding path.

[0093] In one embodiment, when a spine node in the Token computing power network receives data to be distributed and its associated group identifier sent by a leaf node, it uses the group identifier as an index to perform a matching query in the group forwarding table, quickly retrieves the second forwarding path corresponding to the group identifier, and then initiates data packet encapsulation and forwarding operations based on the path.

[0094] Step S4.4: The spine node forwards the data to be distributed and the group identifier to the leaf node pointed to by the second forwarding path.

[0095] In one embodiment, step S4.4 includes steps S4.4.1, S4.4.2, and S4.4.3, which are further described below.

[0096] Step S4.4.1: In response to the fact that the second forwarding path includes multiple forwarding paths, perform a hash function calculation for each second forwarding path.

[0097] When a spine node sends data to be distributed and a group identifier to a leaf node, it does not need to copy them. Specifically, when the second forwarding path is a single forwarding path, the spine node directly forwards the data to be distributed and the group identifier to the leaf node pointed to by the second forwarding path.

[0098] In one embodiment, when the second forwarding path includes multiple forwarding paths, to avoid the network bandwidth waste and data duplication burden caused by simple multipath replication, it is necessary to select one of the multiple forwarding paths, that is, to select and forward according to a hash algorithm. Optionally, key feature fields of the data to be distributed (such as source IP address, destination group identifier, data stream sequence number, or task ID, etc.) can be extracted as input parameters for the hash algorithm, and a hash function is calculated for each second forwarding path to obtain a unique hash digest value corresponding to each path.

[0099] Step S4.4.2: Determine the target forwarding path from the second forwarding paths based on the hash function calculation results of each second forwarding path.

[0100] In one embodiment, a single target forwarding path is determined from multiple second forwarding paths according to a preset path selection strategy, such as selecting the one with the largest or smallest hash value.

[0101] Step S4.4.3: Forward the data to be distributed and the group identifier to the leaf node pointed to by the target forwarding path.

[0102] The data to be distributed and the group identifier are sent to the next-hop node pointed to by the target forwarding path, where the next hop is the leaf node responsible for relaying. Further, the leaf node executes steps S4.1 and S4.2 to forward the data to be distributed and the group identifier again.

[0103] The technical effect of step S4.4 is that it realizes deterministic single-path selection and accurate traffic mapping among multiple alternative paths, effectively avoids the waste of token computing power network resources and the overhead of data deduplication at the receiving end caused by full replication, and improves the transmission efficiency of token computing power network.

[0104] Please see Figure 4 , Figure 4 This invention provides a flowchart for forwarding data to be distributed, such as... Figure 4 As shown, taking a GPU node group consisting of 8 target GPU nodes and a forwarding node consisting of one spine node and two leaf nodes as an example, the 8 target GPU nodes are GPU1 to GPU8, the spine node is spine1, and the two leaf nodes are leaf1 and leaf2. spine1 connects to leaf1 and leaf2, leaf1 connects to GPU1 to GPU4, and leaf2 connects to GPU5 to GPU8. If GPU1 has data to be distributed, it publishes the group identifier and the data to be distributed in the Token computing power network. Leaf1 receives the data to be distributed and the group identifier, looks up the group forwarding table according to the group identifier, obtains 4 first forwarding paths, copies the data to be distributed and the group identifier, and sends them to the GPU nodes (including GPU2, GPU3, GPU4) and the spine node spine1 pointed to by the first forwarding path. Spine1 receives the data to be distributed and the group identifier, looks up the group forwarding table based on the group identifier, obtains the second forwarding path, and forwards the data to be distributed and the group identifier to the leaf node leaf2 pointed to by the second forwarding path; leaf2 receives the data to be distributed and the group identifier, looks up the group forwarding table based on the group identifier, obtains 4 first forwarding paths, copies the data to be distributed and the group identifier and sends them to the GPU nodes (including GPU5~GPU8) pointed to by the first forwarding paths.

[0105] This invention can improve the elasticity of computing power platforms, increase the efficiency of computing power center cluster expansion, enhance the transmission efficiency and speed of token computing network, reduce token output latency, and reduce network congestion.

[0106] Please see Figure 5 , Figure 5 This invention provides a flowchart for updating a group forwarding table, such as... Figure 5 As shown, the method also includes: Step S5: In response to the release of any target GPU node, call the forwarding node to update the group forwarding table to delete the forwarding path corresponding to the released target GPU node.

[0107] When any target GPU node goes offline, the forwarding node updates the group forwarding table and deletes the corresponding forwarding path.

[0108] In one embodiment, the association between the released target GPU node and the group identifier is also removed.

[0109] In one embodiment, when any target GPU node is released due to task completion, failure, or resource reclamation, it indicates that the target GPU node is offline and triggers the group forwarding table update process. The forwarding node is called to retrieve all forwarding path entries in the group forwarding table that involve the released GPU node and delete or mark them as invalid.

[0110] Furthermore, the group member information of the GPU node group is updated synchronously, and the consistency and real-time performance of the group forwarding table data of the Token computing power network are ensured.

[0111] The technical effect of step S5 is to achieve real-time shrinkage of the group forwarding path and immediate isolation of failed nodes, effectively preventing data from being incorrectly forwarded to released GPU nodes, avoiding waste of Token computing power network bandwidth and receiver abnormalities, and improving the transmission accuracy of Token computing power network.

[0112] Step S6: In response to any candidate GPU node joining the GPU node group, establish the association between the candidate GPU node and the group identifier.

[0113] Once any candidate GPU node meets the conditions for joining the GPU node group and passes identity and resource verification, an association is established between the candidate GPU node and the group identifier, a mapping record is generated, and the record is stored. Optionally, the conditions for joining the GPU node group may include at least one of the following: passing computing power authentication, passing health checks, resource matching, and load below a threshold.

[0114] The technical effect of step S6 is to enable the rapid management of new nodes and the instant expansion of group topology in the Token computing power network, ensure that the newly added computing power resources can be integrated into group communication, avoid forwarding errors caused by different data, and improve the accuracy of Token output.

[0115] Step S7: Call the forwarding node to update the group forwarding table to add forwarding paths corresponding to candidate GPU nodes.

[0116] Furthermore, an incremental update notification is triggered, instructing the forwarding node to add the corresponding forwarding path to its local group forwarding table. In other words, when a new GPU node joins, the forwarding node updates its group forwarding table and adds the new forwarding path.

[0117] The technical effect of step S7 is that the new node can receive data traffic within the group immediately after it goes online, without waiting for the periodic group forwarding table to be refreshed; by inserting entries instead of rebuilding the entire table, the impact of the update operation on the Token computing power network forwarding data is minimized, ensuring the continuous stability of the Token computing power network data plane and providing efficient and lossless forwarding plane support for the high-frequency elastic scaling of the computing power platform.

[0118] Please see Figure 6 , Figure 6 This is a structural diagram of a token computing power network forwarding enhancement device for an intelligent computing cloud platform provided by the present invention, as shown below. Figure 6 As shown, the Token computing power network forwarding enhancement device 600 of the intelligent computing cloud platform includes: The node determination module 610 is used to determine the GPU node group to execute the task from the Token computing power network according to the task instructions. The GPU node group includes multiple target GPU nodes. The information generation module 620 is used to generate the group identifier and group forwarding table for the GPU node group; The data publishing module 630 is used to publish the group identifier and the data to be distributed in the Token computing power network for each target GPU node in response to the target GPU node having data to be distributed; The data forwarding module 640 is used to call the forwarding nodes of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

[0119] In one embodiment, the information generation module 620 is further configured to: Generate group identifiers for GPU node groups and establish the association between each target GPU node and the group identifier; Each target GPU node is invoked to publish group identifier indication information on the Token computing power network. The group identifier indication information is used to indicate the group identifier associated with the target GPU node. The system calls forwarding nodes in the computing power network to receive group identifier indication information and obtains the group forwarding table based on the forwarding path of the group identifier information.

[0120] In one embodiment, the information generation module 620 is further configured to: Obtain the source information and interface information of the group identifier; Based on the forwarding path of the group identifier, the source information of the group identifier, and the interface information, obtain the group forwarding table.

[0121] In one embodiment, the forwarding node includes leaf nodes and ridge nodes, and the data forwarding module 640 is further configured to: In response to the leaf node receiving the data to be distributed and the group identifier, the leaf node looks up the group forwarding table based on the group identifier to obtain the first forwarding path; The leaf node sends the data to be distributed and the group identifier to the GPU node or ridge node pointed to by the first forwarding path.

[0122] In one embodiment, the data forwarding module 640 is further configured to: In response to the first forwarding path including multiple forwarding paths, the data to be distributed and the group identifier are copied according to the number of the first forwarding paths; The copied data to be distributed, along with the group identifier, is sent to the GPU node or ridge node pointed to by each first forwarding path.

[0123] In one embodiment, the data forwarding module 640 is further configured to: In response to the spine node receiving the data to be distributed and the group identifier, the spine node looks up the group forwarding table based on the group identifier to obtain the second forwarding path; The spine node forwards the data to be distributed and the group identifier to the leaf node pointed to by the second forwarding path.

[0124] In one embodiment, the data forwarding module 640 is further configured to: Since the second forwarding path includes multiple forwarding paths, a hash function is calculated for each second forwarding path; Based on the hash function calculation results of each second forwarding path, the target forwarding path is determined from the second forwarding paths; Forward the data to be distributed and the group identifier to the leaf node pointed to by the target forwarding path.

[0125] In one embodiment, the token computing power network forwarding enhancement device 600 of the intelligent computing cloud platform further includes an update module 650, used for: In response to the release of any target GPU node, the forwarding node is invoked to update the group forwarding table to delete the forwarding path corresponding to the released target GPU node.

[0126] In one embodiment, the update module 650 is also used for: In response to any candidate GPU node joining the GPU node group, an association is established between the candidate GPU node and the group identifier; The forwarding node is invoked to update the group forwarding table to add forwarding paths corresponding to candidate GPU nodes.

[0127] This application can improve the elasticity of computing power platforms, increase the efficiency of computing power center cluster expansion, optimize network forwarding paths, improve the transmission efficiency of token computing power networks and the speed of token calculation, reduce the latency of token output, and reduce network congestion.

[0128] The Token computing power network forwarding enhancement device for the intelligent computing cloud platform provided by this invention can realize the various processes of the various embodiments of the Token computing power network forwarding enhancement method for the intelligent computing cloud platform. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0129] The Token computing power network forwarding enhancement device for the intelligent computing cloud platform provided by this invention can realize the various processes of the various embodiments of the Token computing power network forwarding enhancement method for the intelligent computing cloud platform. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0130] It should be noted that the Token computing power network forwarding enhancement device of the intelligent computing cloud platform in this invention can be a device, or it can be a component, integrated circuit, or chip in an electronic device.

[0131] The present invention also provides an electronic device, see below. Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 701, a processor 702, and a program or instructions stored in the memory 701 that run on the memory. When the program or instructions are executed by the processor 702, they can achieve the following: Figure 1 The steps in the corresponding intelligent computing cloud platform's token computing power network forwarding enhancement method embodiment, and the achievement of the same beneficial effects, will not be elaborated here.

[0132] The processor 702 can be a CPU, ASIC, FPGA, or GPU.

[0133] Those skilled in the art will understand that all or part of the steps of the above-described embodiment of the Token computing power network forwarding enhancement method for intelligent computing cloud platform can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0134] The present invention also provides a readable storage medium on which a computer program is stored, and which, when executed by a processor, can perform the above-described functions. Figure 1 Any step in the corresponding embodiment of the token computing power network forwarding enhancement method for the intelligent computing cloud platform can achieve the same technical effect, and will not be described again here to avoid repetition. The storage medium mentioned includes, for example, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0135] The present invention also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described... Figure 1 The various processes of the corresponding intelligent computing cloud platform's token computing power network forwarding enhancement method embodiment can achieve the same technical effect, and will not be described again here to avoid repetition.

[0136] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Additionally, the use of "and / or" in this invention indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: A alone, B alone, C alone, both A and B present, both B and C present, both A and C present, and A, B, and C present.

[0137] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or second terminal device, etc.) to execute the methods of the various embodiments of the present invention.

[0139] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A method for enhancing token computing power network forwarding in an intelligent computing cloud platform, characterized in that, include: Step S1: According to the task instructions, determine the GPU node group to execute the task from the Token computing power network, wherein the GPU node group includes multiple target GPU nodes; Step S2: Generate the group identifier and group forwarding table for the GPU node group; Step S3: For each target GPU node, in response to the target GPU node having data to be distributed, publish the group identifier and the data to be distributed in the Token computing power network; Step S4: Invoke the forwarding node of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

2. The method according to claim 1, characterized in that, Step S2 includes: Step S2.1: Generate a group identifier for the GPU node group and establish an association between each target GPU node and the group identifier; Step S2.2: Invoke each of the target GPU nodes to publish group identifier indication information in the Token computing power network. The group identifier indication information is used to indicate the group identifier associated with the target GPU node. Step S2.3: Call the forwarding node in the computing power network to receive the group identifier indication information, and obtain the group forwarding table according to the forwarding path of the group identifier information.

3. The method according to claim 2, characterized in that, Step S2.3 includes: Step S2.3.1: Obtain the source information and interface information of the group identifier; Step S2.3.2: Obtain the group forwarding table based on the forwarding path of the group identifier information, the source information of the group identifier, and the interface information.

4. The method according to any one of claims 1-3, characterized in that, The forwarding node includes leaf nodes and spine nodes, and step S4 includes: Step S4.1: In response to the leaf node receiving the data to be distributed and the group identifier, the leaf node searches the group forwarding table based on the group identifier to obtain the first forwarding path; Step S4.2: The leaf node sends the data to be distributed and the group identifier to the GPU node or ridge node pointed to by the first forwarding path.

5. The method according to claim 4, characterized in that, Step S4.2 includes: Step S4.2.1: In response to the first forwarding path including multiple forwarding paths, the data to be distributed and the group identifier are copied according to the number of the first forwarding paths; Step S4.2.2: Send the copied data to be distributed and the group identifier to the GPU node or spine node pointed to by each of the first forwarding paths.

6. The method according to claim 4, characterized in that, Also includes: Step S4.3: In response to the spine node receiving the data to be distributed and the group identifier, the spine node searches the group forwarding table based on the group identifier to obtain the second forwarding path; Step S4.4: The spine node forwards the data to be distributed and the group identifier to the leaf node pointed to by the second forwarding path.

7. The method according to claim 6, characterized in that, Step S4.4 includes: Step S4.4.1: In response to the fact that the second forwarding path includes multiple forwarding paths, a hash function is calculated for each of the second forwarding paths; Step S4.4.2: Determine the target forwarding path from the second forwarding paths based on the hash function calculation results of each second forwarding path; Step S4.4.3: Forward the data to be distributed and the group identifier to the leaf node pointed to by the target forwarding path.

8. The method according to any one of claims 1-3, characterized in that, Also includes: Step S5: In response to the release of any of the target GPU nodes, the forwarding node is invoked to update the group forwarding table to delete the forwarding path corresponding to the released target GPU node.

9. The method according to any one of claims 1-3, characterized in that, Also includes: Step S6: In response to any candidate GPU node joining the GPU node group, establish an association between the candidate GPU node and the group identifier; Step S7: Call the forwarding node to update the group forwarding table to add the forwarding path corresponding to the candidate GPU node.

10. A token computing power network forwarding enhancement device for an intelligent computing cloud platform, characterized in that, include: The node determination module is used to determine the GPU node group to execute the task from the Token computing power network according to the task instructions, wherein the GPU node group includes multiple target GPU nodes; The information generation module is used to generate the group identifier and group forwarding table of the GPU node group; The data publishing module is used to publish the group identifier and the data to be distributed in the Token computing power network for each target GPU node in response to the target GPU node having data to be distributed; The data forwarding module is used to call the forwarding nodes of the Token computing power network to forward the data to be distributed according to the group identifier and the group forwarding table.

11. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the Token computing power network forwarding enhancement method for the intelligent computing cloud platform as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the Token computing power network forwarding enhancement method for the intelligent computing cloud platform as described in any one of claims 1 to 9.

13. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the steps of the Token computing power network forwarding enhancement method for the intelligent computing cloud platform as described in any one of claims 1 to 9.

Images