Data transmission methods, devices, storage media, and electronic equipment
By configuring different groups of transmission paths in the data center network, which then pass through different switches, the self-congestion problem caused by data flow convergence in cross-regional communication is solved, and data transmission efficiency is improved.
Patent Information
- Application Number
- CN202310481723.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-27
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-04-27
AI Technical Summary
In data transmission scenarios, especially in cross-regional communication, data streams converging in the same switch can cause self-congestion, resulting in low data transmission efficiency, a problem that existing technologies have not been able to effectively solve.
By identifying the target server pair from multiple servers in the data center network, obtaining the data flow transmission path information, and configuring at least two sets of transmission paths based on this information, the data flow is guided to use different sets of transmission paths for transmission through different switches.
This reduces the probability of data streams converging on the same switch, improves data transmission efficiency, and solves the problem of low data transmission efficiency.
Smart Images

Figure CN118869604B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computers, and more specifically, to a data transmission method, apparatus, storage medium, and electronic device. Background Technology
[0002] In data transmission scenarios, the interaction between the sender and receiver is usually achieved through switches. However, if a large amount of data flow converges in the same switch, or if different switches do not handle the data flow evenly, there is a certain probability that self-congestion and other problems will occur, which will lead to low data transmission efficiency.
[0003] Furthermore, when the sending and receiving ends communicate across regions, the probability of such large data flows converging in the same switch is higher, thus increasing the likelihood of congestion and other problems. Therefore, data transmission efficiency is relatively low.
[0004] There is currently no effective solution to the above problems. Summary of the Invention
[0005] This application provides a data transmission method, apparatus, storage medium, and electronic device to at least solve the technical problem of low data transmission efficiency.
[0006] According to one aspect of the embodiments of this application, a data transmission method is provided, comprising: determining a target server pair from a plurality of servers in a data center network, wherein the data center network further includes a plurality of target switches, and the target server pair consists of two servers communicating across regions; acquiring data path information during the transmission of a data stream by the target server pair, wherein the data path information is used to indicate the target switches through which the data stream passes during the transmission process; configuring at least two sets of transmission paths based on the data path information, wherein transmission paths in the same group of the at least two sets of transmission paths pass through the same target switches, and transmission paths in different groups of the at least two sets of transmission paths pass through different target switches; in response to a data transmission request triggered on the target server pair, configuring at least two transmission paths for the target server pair based on the different groups of transmission paths, and instructing the target server pair to transmit a target data stream through the at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0007] According to another aspect of the embodiments of this application, a data transmission apparatus is also provided, comprising: a determining unit, configured to determine a target server pair from a plurality of servers in a data center network, wherein the data center network further includes a plurality of target switches, and the target server pair consists of two servers communicating across regions; a first acquiring unit, configured to acquire data path information during the transmission of a data stream by the target server pair, wherein the data path information is used to indicate the target switches through which the data stream passes during the transmission process; a configuring unit, configured to configure at least two sets of transmission paths based on the data path information, wherein transmission paths in the same set of the at least two sets of transmission paths pass through the same target switches, and transmission paths in different sets of the at least two sets of transmission paths pass through different target switches; and a transmission unit, configured to, in response to a data transmission request triggered on the target server pair, configure at least two transmission paths for the target server pair based on the different sets of transmission paths, and instruct the target server pair to transmit a target data stream through the at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0008] As an optional solution, the above-mentioned device further includes: a second acquisition unit, configured to acquire N probe data streams before acquiring the data path information during the data stream transmission process of the target server, wherein each probe data stream in the plurality of probe data streams is configured with a different source port number, the different source port number being used to indicate different transmission paths of the probe data streams in the data center network, and N being an integer greater than 1; a control unit, configured to control the target server to transmit the N probe data streams before acquiring the data path information during the data stream transmission process of the target server; the first acquisition unit includes: a first acquisition module, configured to acquire probe data path information during the transmission process of the N probe data streams, wherein the probe data path information being used to indicate the target switch through which the probe data streams pass during the transmission process of the N probe data streams.
[0009] As an optional solution, the above configuration unit includes: an execution module, configured to perform the following steps until at least two sets of transmission paths are obtained: acquiring current data path information, configuring at least two sets of candidate paths based on the current data path information, and acquiring the number of current candidate paths in each set of candidate paths; if the number of current candidate paths is less than a preset number, acquiring M probe data streams, wherein the M probe data streams are the next batch of probe data streams of the N probe data streams; controlling the target server to transmit the M probe data streams; acquiring probe data path information during the transmission of the M probe data streams, and using the probe data path information during the transmission of the M probe data streams as the current data path information, configuring at least two sets of candidate paths based on the current data path information; if the number of current candidate paths is greater than or equal to the preset number, determining the at least two sets of candidate paths as the at least two sets of transmission paths.
[0010] As an optional solution, the above configuration unit includes: a grouping module, used to group the source port numbers set by the above data path information to obtain at least two groups of source port numbers, wherein the source port numbers of the same group in the at least two groups of source port numbers indicate the target switches through which the above probe data streams pass in the same transmission path in the above data center network, and the source port numbers of the same group in the at least two groups of source port numbers indicate the target switches through which the above probe data streams pass in the transmission path in the above data center network.
[0011] As an optional solution, the above configuration unit includes: a second acquisition module, used to acquire a first transmission path through the first switch and a second transmission path through the second switch based on the above data path information, wherein the first switch and the second switch are different target switches among the plurality of target switches; and a determination module, used to determine the first transmission path and the second transmission path as transmission paths of different groups.
[0012] As an optional solution, the first acquisition unit includes: a third acquisition module, used to acquire entry information associated with the receiving server, wherein the receiving server is the server acting as the receiving end in the target server pair, the data path information includes the entry information, the entry information is used to instruct the first network port or the second network port on the receiving server to receive the current data stream, the first network port is connected to the first switch, and the second network port is connected to the second switch; the second acquisition module includes: a determining submodule, used to, when the entry information instructs the first network port to receive the current data stream, use the transmission path of the current data stream in the data center network as the first transmission path; or, when the entry information instructs the second network port to receive the current data stream, use the transmission path of the current data stream in the data center network as the second transmission path.
[0013] As an optional solution, the above-mentioned device further includes: a fourth acquisition module, used to acquire the source port number set by the current data stream before acquiring the first transmission path of the first switch and the second transmission path of the second switch based on the data path information, wherein the source port number set by the current data stream is used to indicate the transmission path of the current data stream in the data center network.
[0014] As an optional solution, the transmission unit includes: a first determining module, configured to determine the different groups of transmission paths of the specified number of paths from the at least two sets of transmission paths when the number of paths of the at least two transmission paths is less than the number of path groups of the at least two sets of transmission paths; or, a second determining module, configured to obtain the same number of transmission paths from each group of transmission paths of the at least two sets of transmission paths when the number of paths of the at least two transmission paths is equal to the number of path groups of the at least two sets of transmission paths; or, a third determining module, configured to obtain the same or different number of transmission paths from each group of transmission paths when the number of paths of the at least two transmission paths is greater than the number of path groups of the at least two sets of transmission paths, wherein the difference in the number of transmission paths obtained from each group of transmission paths is less than a preset threshold.
[0015] As an optional solution, the aforementioned transmission unit includes: a transmission module, configured to, in response to a collaborative training request triggered by the target server pair, configure at least two transmission paths for the target server pair based on the transmission paths of the different groups, and instruct the target server pair to transmit training data streams through the at least two transmission paths, wherein the collaborative training request is used to request the target server pair to perform collaborative training through the transmission of the training data streams.
[0016] According to another aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the data transmission method described above.
[0017] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described data transmission method through the computer program.
[0018] In this embodiment, a target server pair is determined from multiple servers in a data center network, wherein the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions. Data path information is obtained during the data flow transmission process of the target server pair, wherein the data path information indicates the target switches through which the data flow passes during the transmission process. At least two sets of transmission paths are configured based on the data path information, wherein transmission paths in the same group of the at least two sets of transmission paths pass through the same target switches, and transmission paths in different groups of the at least two sets of transmission paths pass through different target switches. In response to a data transmission request triggered on the target server pair, at least two transmission paths are configured for the target server pair based on the different groups of transmission paths, and the target server pair is instructed to transmit the target data flow through the at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data flow.
[0019] This embodiment pre-groups the transmission paths of data streams according to switches with different pathways. Furthermore, during the transmission process, it guides the data streams to use different groups of transmission paths, thereby reducing the probability of a large number of data streams clustering on the same switch. This achieves the technical effect of improving data transmission efficiency and solves the technical problem of low data transmission efficiency. Attached Figure Description
[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:
[0021] Figure 1 This is a schematic diagram of an application environment for an optional data transmission method according to an embodiment of this application;
[0022] Figure 2 This is a schematic diagram of the flow of an optional data transmission method according to an embodiment of this application;
[0023] Figure 3 This is a schematic diagram of an optional data transmission method according to an embodiment of this application;
[0024] Figure 4 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0025] Figure 5 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0026] Figure 6 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0027] Figure 7 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0028] Figure 8 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0029] Figure 9 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0030] Figure 10 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0031] Figure 11 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0032] Figure 12 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0033] Figure 13 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0034] Figure 14 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0035] Figure 15 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0036] Figure 16 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0037] Figure 17 This is a schematic diagram of another optional data transmission method according to an embodiment of this application;
[0038] Figure 18 This is a schematic diagram of an optional data transmission device according to an embodiment of this application;
[0039] Figure 19 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application. Detailed Implementation
[0040] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.
[0041] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. 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 such processes, methods, products, or apparatus.
[0042] For ease of understanding, the following terms are explained:
[0043] Cloud computing is a computing model that distributes computing tasks across a large pool of computers, enabling various application systems to access computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." From the user's perspective, resources in the "cloud" appear infinitely scalable, readily available, on-demand, and expandable, with payment based on usage.
[0044] As a provider of fundamental cloud computing capabilities, a cloud resource pool (referred to as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform) is established. Various types of virtual resources are deployed in the resource pool for external customers to choose from. The cloud resource pool mainly includes: computing devices (virtualized machines containing operating systems), storage devices, and network devices.
[0045] Based on logical function, a PaaS (Platform as a Service) layer can be deployed on top of the IaaS (Infrastructure as a Service) layer, and a SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Alternatively, SaaS can be deployed directly on top of IaaS. PaaS is a platform for running software, such as databases and web containers. SaaS refers to various types of business software, such as web portals and bulk SMS senders. Generally speaking, SaaS and PaaS are upper layers compared to IaaS.
[0046] According to one aspect of the embodiments of this application, a data transmission method is provided. Optionally, as an optional implementation, the above-described data transmission method may be applied to, but is not limited to, [examples of other methods]. Figure 1 The environment shown may include, but is not limited to, user equipment 102 and server 112. User equipment 102 may include, but is not limited to, a display 104, a processor 106 and a memory 108. Server 112 includes a database 114 and a processing engine 116.
[0047] The specific process can be summarized in the following steps:
[0048] Step S101: Server 112 determines the target server pair from multiple servers in the data center network; obtains data path information during the data flow transmission process of the target server pair; and configures at least two sets of transmission paths based on the data path information.
[0049] In step S102, user equipment 102 obtains a data transmission request triggered by a target server among multiple servers;
[0050] Steps S104-S106: Send the data transmission request to the server 112 via network 110;
[0051] In step S108, server 112 configures at least two transmission paths for target server pair based on transmission paths of different groups through processing engine 116, and instructs target server pair to transmit target data stream through at least two transmission paths to obtain data transmission result.
[0052] In steps S110-S112, the data transmission result is sent to the user equipment 102 via the network 110. The user equipment 102 displays the data transmission result on the display 104 via the processor 106 and stores the data transmission result in the memory 108.
[0053] remove Figure 1 Beyond the examples shown, the above steps can be performed independently by the user equipment or the server, or collaboratively by both, such as by user equipment 102 executing steps S101, S108, etc., thereby reducing the processing load on server 112. User equipment 102 includes, but is not limited to, handheld devices (such as mobile phones), laptops, tablets, desktop computers, in-vehicle devices, smart TVs, etc. This application does not limit the specific implementation of user equipment 102. Server 112 can be a single server, a server cluster consisting of multiple servers, or a cloud server.
[0054] Alternatively, as an alternative implementation method, such as Figure 2 As shown, the data transmission method can be performed by an electronic device, such as... Figure 1 The user equipment or server shown includes the following specific steps:
[0055] S202, Identify a target server pair from multiple servers in a data center network, where the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions;
[0056] S204, Obtain data path information during the transmission of data streams by the target server, wherein the data path information is used to indicate the target switch through which the data stream passes during transmission.
[0057] S206, Configure at least two sets of transmission paths based on data path information, wherein the transmission paths of the same group in the at least two sets of transmission paths have the same target switch, and the transmission paths of different groups in the at least two sets of transmission paths have different target switches.
[0058] S208, in response to a data transmission request triggered by a target server pair, configure at least two transmission paths for the target server pair based on transmission paths of different groups, and instruct the target server pair to transmit the target data stream through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0059] Optionally, in this embodiment, the above data transmission method can be applied to, but is not limited to, scenarios related to data center networks. Specifically, it can be applied to scenarios involving large-scale data transmission between multiple servers in a data center network, such as artificial intelligence (AI) model training scenarios. For the data center network, a Clos / Fattree architecture (layer three or layer two switches) can be used, but is not limited to, such as... Figure 3 As shown, there are three layers of switches: access layer (Leaf), aggregation layer (Spine), and core layer (Core), and one layer of servers (Host). The access layer switches and their downstream servers together form a rack. The access layer switches, all their upstream aggregation layer switches, and all their downstream servers together form a pod. Further... Figure 3 In the diagram, access switches L0 and L1 and their downstream servers H0 and H1 form a rack; while all the devices within the first dashed box form a pod.
[0060] It should be noted that, to increase communication bandwidth and connection reliability between servers, servers can, but are not limited to, use two links connected to two access layer switches (one network card with two network ports connected to two switches respectively). In data center networks, such as... Figure 4 As shown, the communication topology between a pair of servers can be roughly divided into three categories: within the Rack (via Leaf switches), across the Rack (via Spine switches), and across Pods (via Core switches). Due to the existence of aggregated links with dual uplinks on server nodes and the randomness of ECMP hash routing, the latter two types of communication will face the problem of throughput not fully utilizing the capacity of the two links (self-congestion problem). Figure 5 As shown, assuming each uplink has a bandwidth capacity of 100Gbps (the total bandwidth after link aggregation is 200Gbps, but in reality, the link bandwidth can be other values, including but not limited to 10G, 25G, 200G, etc.), when the server uses 2 connections, there is a 50% probability that the throughput will only be 100Gbps (case 1 and case 4). When more connections are used, the traffic distribution becomes more complex; the relationship between average total throughput and the number of connections is shown in [reference needed]. Figure 6 As shown, before the number of connections reaches 10, the average total throughput of the unoptimized solution (the existing pure ECMP routing-based solution) cannot reach 200G. Furthermore, congestion control in the network interface cards of high-performance AI training clusters in related technologies is based on IP granularity (i.e., all data streams destined for the same IP share the same rate). In this situation, the combined effect of load imbalance leads to a decrease in the average throughput of cross-rack communication.
[0061] Furthermore, regarding the relationship between AI model training and data center networks, training can be conducted using, but is not limited to, high-performance computing (HPC) clusters composed of distributed servers within the data center network. In such computing clusters, multiple servers equipped with high-performance GPUs collaborate for training via high-bandwidth networks (such as RoCEv2 or IB networks). However, based on this hardware, parameter server (PS) training architectures are difficult to scale to large-scale networks. Peer-to-peer architectures, which have emerged to address this issue, can effectively solve the scalability problem. In peer-to-peer architectures, training primarily relies on Ring Allreduce ensemble operations to complete data communication. Ring Allreduce uses ring topology and pipeline techniques for parallel transmission, thereby minimizing training latency and improving network linearity.
[0062] However, using ring topologies and pipeline techniques for parallel transmission often suffers from the reduced average throughput of cross-region (e.g., Rack / Pod) communication, as seen in the ring topology described above, where the overall cluster throughput is constrained by the minimum throughput communication link. Figure 6 As shown, under optimal configuration, this 8-node cluster has only 2 sets of cross-rack communication traffic. If communication between H3 and H4 experiences low throughput, the effective throughput of the entire cluster will drop to 100G, as... Figure 7 As shown. This is still the optimal scenario; more generally, training clusters may experience more cross-rack communication, primarily for the following two reasons:
[0063] First, the communication topology is not optimized. Aggregate communication typically generates a topology loop using a list of server nodes provided to it. If the order of the node list is optimized, such as... Figure 8 The communication topology shown is generated and optimized from the node list [H0, H4, H1, H5, H2, H6, H3, H7]. This results in less cross-rack communication traffic, generating only 4 cross-rack communication groups. However, if the node list order is not optimized, such as... Figure 9 The communication topology in the example is generated from the node list [H0, H4, H1, H5, H2, H6, H3, H7] without optimization. This results in more cross-rack communication traffic, leading to 8 cross-rack communication groups.
[0064] Secondly, the cluster size is very large, with servers distributed across multiple racks. When the cluster size is large, it's inevitable to distribute a large number of servers across multiple racks or even multiple pods, resulting in a significant amount of cross-rack communication. Furthermore, simple mathematical probability analysis shows that the more cross-rack communication occurs, the greater the probability of low-throughput communication, and consequently, the greater the probability of low-throughput cluster performance, making it easier to reduce data transmission efficiency. For example, in a large cluster, the effective throughput of the entire cluster is likely only 100G, which is about 50% of the ideal throughput.
[0065] Furthermore, this embodiment optimizes the above-mentioned issue of reduced average throughput in cross-regional communication to ensure that traffic on aggregated links in cross-rack communication does not experience self-congestion (e.g., Figure 5 This involves cases 2 and 3, ensuring that the effective throughput of the AI training cluster reaches the optimal level and maximizing the efficiency of the training cluster.
[0066] Optionally, in this embodiment, the target switch may be, but is not limited to, different switch types, such as a target switch belonging to the aggregation layer, a target switch belonging to the access layer, a target switch belonging to the core layer, or a target switch belonging to the aggregation layer, etc.
[0067] To further illustrate, optional examples include... Figure 4 As shown, the switches belonging to the core layer include switch C0 and switch C2; the switches belonging to the aggregation layer include switch S0, switch S1, switch S4, and switch S5; and the switches belonging to the access layer include switch L0, switch L1, switch L2, switch L3, switch L4, and switch L5.
[0068] Optionally, in this embodiment, the target server pair refers to two servers communicating across regions. This embodiment considers that when communicating across regions, the probability of a large amount of data converging on the same switch is higher, thus increasing the probability of congestion and other problems. Therefore, path configuration is specifically performed for server pairs communicating across regions to reduce the negative impact of cross-regional communication and reduce the probability of a large amount of data converging on the same switch. Furthermore, the target server pair can also be, but is not limited to, two servers communicating across regions.
[0069] Optionally, in this embodiment, the data path information is used to indicate the target switch through which the data stream passes during transmission. The acquisition of the data path information can be divided into direct acquisition or indirect acquisition, but is not limited to. For direct acquisition, it can be, but is not limited to, directly acquiring the switch ID through which the data stream passes during transmission, and thus determining the target switch through which the data stream passes during transmission. For indirect acquisition, it can be, but is not limited to, using multiple network ports set on the receiving end and connected one by one to the target switch, and obtaining which network port received the data stream, thus indirectly determining which target switch the data stream passed through during transmission.
[0070] Optionally, in this embodiment, transmission paths within the same group of at least two transmission paths pass through the same target switch, while transmission paths between different groups of at least two transmission paths pass through different target switches. For example, transmission paths passing through the same target switch are grouped into the same path group, thus enabling transmission paths from different groups to pass through different target switches. Since ensuring that transmission paths from different groups pass through different target switches is sufficient, there is no requirement to specify which target switch is being used, thereby avoiding the use of unnecessary resources to determine the target switch and improving the overall efficiency of data transmission.
[0071] Optionally, in this embodiment, at least two transmission paths are configured for the target server pair based on transmission paths from different groups. The at least two transmission paths may, but are not limited to, avoid target switches that share the same path. However, when the number of transmission paths is greater than the number of path groups, it is usually impossible to avoid target switches that share the same path. But the number of transmission paths that share the same path with target switches can be reduced, such as by alternately selecting transmission paths from different groups of path groups as the transmission paths configured for the target server pair.
[0072] It should be noted that by pre-grouping the transmission paths of data streams according to different switches, and further guiding the data streams to use different groups of transmission paths during the transmission process, the probability of a large number of data streams gathering on the same switch is reduced, thereby achieving the technical effect of improving the overall efficiency of data transmission.
[0073] To further illustrate, optional examples include... Figure 10As shown, a target server pair (e.g., server X2 and server Y2) is determined from multiple servers in a data center network 1002. The data center network 1002 also includes multiple target switches 1004 (e.g., switch B1, switch B2, switch Bm, etc.). The target server pair consists of two servers communicating across regions. Data path information is obtained during the data flow transmission process of the target server pair. This data path information indicates the target switches 1004 through which the data flow passes during transmission. At least two sets of transmission paths are configured based on the data path information. The transmission paths in the same group of the at least two sets of transmission paths are... The target switches 1004 with the same path (e.g., both path to switch B2) and the target switches 1004 with different paths between different groups of transmission paths (e.g., transmission path 1 path to switch B1 and transmission path 2 path to switch B2, then transmission path 1 and transmission path 2 are determined to be transmission paths of different groups); in response to a data transmission request triggered by a target server pair, at least two transmission paths are configured for the target server pair based on the transmission paths of different groups, and the target server pair is instructed to transmit the target data stream through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0074] The embodiments provided in this application identify a target server pair from multiple servers in a data center network, wherein the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions. Data path information is obtained during the data flow transmission process of the target server pair, wherein the data path information indicates the target switches through which the data flow passes during transmission. At least two sets of transmission paths are configured based on the data path information, wherein transmission paths in the same group of at least two sets of transmission paths pass through the same target switches, and transmission paths in different groups of at least two sets of transmission paths pass through different target switches. In response to a data transmission request triggered on the target server pair, at least two transmission paths are configured for the target server pair based on the different groups of transmission paths, and the target server pair is instructed to transmit the target data flow through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data flow. By pre-grouping the data flow transmission paths according to the switches with different paths, and further guiding the data flow to use different groups of transmission paths during transmission, the probability of a large number of data flows converging on the same switch is reduced, thereby achieving the technical effect of improving data transmission efficiency.
[0075] As an optional approach, before obtaining the data path information during the data stream transmission process from the target server, the method further includes:
[0076] S1-1, Obtain N probe data streams, where each probe data stream in the multiple probe data streams is set with a different source port number. The different source port numbers are used to indicate different transmission paths of the probe data streams in the data center network. N is an integer greater than 1.
[0077] S1-2, control the target server to transmit N probe data streams;
[0078] Obtaining data path information during the transmission of data streams to the target server includes: obtaining probe data path information during the transmission of N probe data streams, wherein the probe data path information is used to indicate the target switch through which the probe data streams pass during the transmission of the N probe data streams.
[0079] Optionally, in this embodiment, the probe data stream can be used, but is not limited to, to test whether different transmission paths in a data center network pass through the same switch. Furthermore, since four of the five tuples in a routing hash (destination port number, source / destination IP, protocol number) cannot be arbitrarily modified, only the source port number can be arbitrarily modified. Therefore, in this embodiment, by modifying the source port number set in the data stream, the transmission path of the probe data stream in the data center network is indicated, and different source port numbers are set for each probe data stream to indicate different transmission paths in the data center network.
[0080] To further illustrate, the optional based on Figure 10 The scenario shown continues, for example... Figure 11 As shown, the solid line corresponds to the first source port number, the dashed line corresponds to the second source port number, and the bold solid line corresponds to the third source port number.
[0081] The embodiments provided in this application acquire N probe data streams, each of which is configured with a different source port number. These different source port numbers indicate different transmission paths of the probe data streams in the data center network, where N is an integer greater than 1. The target server is controlled to transmit the N probe data streams. Probe data path information is acquired during the transmission of the N probe data streams. This probe data path information indicates the target switches that the probe data streams pass through during the transmission of the N probe data streams. This achieves the purpose of testing whether different transmission paths in the data center network pass through the same switches, thereby improving the technical effect of acquiring data path information.
[0082] As an optional approach, at least two sets of transmission paths are configured based on data path information, including:
[0083] Perform the following steps until at least two transmission paths are obtained:
[0084] S2-1, Obtain the current data path information, configure at least two candidate paths based on the current data path information, and obtain the number of current candidate paths in each of the at least two candidate paths;
[0085] S2-2, when the number of current candidate paths is less than the preset number, obtain M probe data streams, where the M probe data streams are the next batch of probe data streams from the N probe data streams;
[0086] S2-3, control the target server to transmit M probe data streams;
[0087] S2-4, Obtain the probe data path information during the transmission process of M probe data streams, and use the probe data path information during the transmission process of M probe data streams as the current data path information, and configure at least two candidate paths based on the current data path information.
[0088] S2-5, if the current number of candidate paths is greater than or equal to the preset number, determine at least two groups of candidate paths as at least two groups of transmission paths.
[0089] It should be noted that if the number of transmission paths in at least two transmission paths is small, the probability of assigning transmission paths from the same group (passing through the same target switch) to the target server increases, thus affecting the stability of efficient data transmission. To avoid this situation, this embodiment sets a preset number to limit the number of transmission paths in at least two transmission paths.
[0090] To illustrate further, we can assume that the preset number is 10, and the number of transmission paths for group 1 is 10, while the number of transmission paths for group 2 is 9. Obviously, the number of transmission paths for group 2 is less than the preset number. Therefore, the next batch of probe data streams is used for path configuration until the number of transmission paths for group 2 reaches the preset number (10), so as to ensure the number of transmission paths in at least two groups of transmission paths.
[0091] The embodiments provided in this application involve performing the following steps until at least two sets of transmission paths are obtained: acquiring current data path information, configuring at least two sets of candidate paths based on the current data path information, and acquiring the number of current candidate paths in each set of candidate paths; if the number of current candidate paths is less than a preset number, acquiring M probe data streams, wherein the M probe data streams are the next batch of probe data streams after the N probe data streams; controlling the target server to transmit the M probe data streams; acquiring probe data path information during the transmission of the M probe data streams, and using the probe data path information during the transmission of the M probe data streams as the current data path information, configuring at least two sets of candidate paths based on the current data path information; if the number of current candidate paths is greater than or equal to a preset number, determining the at least two sets of candidate paths as at least two sets of transmission paths, thereby achieving the goal of ensuring the number of transmission paths in at least two sets of transmission paths, and thus realizing the technical effect of improving the stability of efficient data transmission.
[0092] As an optional approach, at least two sets of transmission paths are configured based on data path information, including:
[0093] The source port numbers of multiple probe data streams are grouped using data path information to obtain at least two groups of source port numbers. Among these, the probe data streams indicated by the same group of source port numbers in the at least two groups of source port numbers pass through the same target switch in the data center network, while the probe data streams indicated by the same group of source port numbers in the at least two groups of source port numbers pass through different target switches in the data center network.
[0094] It should be noted that, for the convenience of path grouping, the transmission paths corresponding to each probe data stream are grouped according to the first target switch with different paths, using the source port number as the number.
[0095] To further illustrate, optional examples include... Figure 11 As shown, the source port number of the transmission path corresponding to the solid line is then used as the first source port number in group 1, the source port number of the transmission path corresponding to the dashed line is used as the first source port number in group 2, and the source port number of the transmission path corresponding to the bold solid line is used as the second source port number in group 2.
[0096] The embodiments provided in this application group the source port numbers of multiple probe data streams using data path information to obtain at least two groups of source port numbers. Among these, the probe data streams indicated by the same group of source port numbers in the at least two groups of source port numbers pass through the same target switch in the data center network, while the probe data streams indicated by the same group of source port numbers in the at least two groups of source port numbers pass through different target switches in the data center network. This achieves the purpose of grouping the transmission paths corresponding to each probe data stream according to the first target switch with different paths, using the source port number as the number, thereby improving the technical effect of improving path grouping efficiency.
[0097] As an optional approach, at least two sets of transmission paths are configured based on data path information, including:
[0098] S3-1, based on data path information, obtain the first transmission path of the first switch and the second transmission path of the second switch, wherein the first switch and the second switch are different first target switches among multiple first target switches;
[0099] S3-2, the first transmission path and the second transmission path are determined to be transmission paths of different groups.
[0100] It should be noted that, in order to facilitate the configuration of at least two sets of transmission paths, transmission paths that pass through different switches are determined as different sets of transmission paths.
[0101] To further illustrate, optional examples include... Figure 11 As shown, dashed lines, bold solid lines, and solid lines represent three transmission paths. Further assuming that switch B1 is the first switch and switch B2 is the second switch, the first transmission path through the first switch includes the transmission path corresponding to the solid line, and the second transmission path through the second switch includes the transmission paths corresponding to the dashed lines and bold solid lines.
[0102] Through the embodiments provided in this application, a first transmission path through a first switch and a second transmission path through a second switch are obtained based on data path information, wherein the first switch and the second switch are different first target switches among a plurality of first target switches; the first transmission path and the second transmission path are determined as transmission paths of different groups, thereby achieving the purpose of improving the configuration efficiency of the transmission path, and thus realizing the technical effect of improving data transmission efficiency.
[0103] As an optional solution, the data path information during the transmission of data streams between the target servers is obtained, including: obtaining the entry information associated with the receiving server, wherein the receiving server is the server that acts as the receiving end in the target server pair, and the data path information includes entry information, which is used to indicate the first network port or the second network port on the receiving server to receive the current data stream, wherein the first network port is connected to the first switch, and the second network port is connected to the second switch.
[0104] The method of obtaining the first transmission path of the first switch and the second transmission path of the second switch based on the data path information includes: when the ingress information is used to instruct the first network port to receive the current data stream, the transmission path of the current data stream in the data center network is taken as the first transmission path; or, when the ingress information is used to instruct the second network port to receive the current data stream, the transmission path of the current data stream in the data center network is taken as the second transmission path.
[0105] It should be noted that, in order to improve the efficiency of obtaining data path information, an indirect acquisition method is adopted, which uses the first or second network port of the receiving server to receive the current data stream in order to determine which switch the current data stream passes through.
[0106] To further illustrate, optional examples include... Figure 12 As shown, the receiving server 1202 is equipped with network port 1 and network port 2. Network port 1 is connected to switch 1, and network port 2 is connected to switch 2. Therefore, when a data stream is received on network port 1, it can be determined that the data stream has passed through switch 1. Similarly, when a data stream is received on network port 2, it can be determined that the data stream has passed through switch 2. This achieves the goal of indirectly obtaining the switch through which the data stream has passed without having to detect the switch through which the data stream has passed.
[0107] The embodiments provided in this application obtain entry information associated with a receiving server. The receiving server is the server acting as the receiving end in a target server pair. Data path information includes entry information, which instructs either a first network port or a second network port on the receiving server to receive the current data stream. The first network port is connected to a first switch, and the second network port is connected to a second switch. When the entry information instructs the first network port to receive the current data stream, the transmission path of the current data stream in the data center network is used as the first transmission path. Alternatively, when the entry information instructs the second network port to receive the current data stream, the transmission path of the current data stream in the data center network is used as the second transmission path. This achieves the goal of indirectly obtaining the data stream path via a network port without needing to detect the switch in the data stream path, thereby improving the efficiency of obtaining data path information.
[0108] As an optional approach, before obtaining the first transmission path through the first switch and the second transmission path through the second switch based on the data path information, the method further includes:
[0109] Get the source port number set for the current data stream, which indicates the transmission path of the current data stream in the data center network.
[0110] Optionally, since the four tuples (destination port number, source / destination IP, protocol number) in the five-tuple of the routing hash cannot be modified arbitrarily, only the source port number can be modified arbitrarily. Therefore, in this embodiment, the transmission path of the current data flow in the data center network is indicated by modifying the source port number set in the data flow.
[0111] As an optional approach, at least two transmission paths are configured for the target server based on different groups of transmission paths, including:
[0112] S4-1, when the number of paths in at least two transmission paths is less than the number of path groups in at least two sets of transmission paths, determine the transmission paths with different numbers of paths from the at least two sets of transmission paths; or,
[0113] S4-2, where the number of paths in at least two transmission paths equals the number of path groups in at least two sets of transmission paths, obtain the same number of transmission paths from each set of transmission paths in the at least two sets of transmission paths; or,
[0114] S4-3, when the number of paths in at least two transmission paths is greater than the number of path groups in at least two transmission paths, the same or different number of transmission paths are obtained from each group of transmission paths, wherein the difference in the number of transmission paths obtained from each group of transmission paths is less than a preset threshold.
[0115] Optionally, in this embodiment, it is assumed that the target server needs to configure two transmission paths for a, and the number of path groups of at least two transmission paths is 3. Then, any two transmission paths can be selected from the at least two transmission paths, and one transmission path can be determined from each of the two transmission paths, thereby obtaining two transmission paths with different groups, and configuring them for the target server a.
[0116] Optionally, in this embodiment, it is assumed that the target server needs to configure 3 transmission paths for a, and the number of path groups of at least two transmission paths is 3. Then, one transmission path is determined from each of the 3 transmission paths, resulting in 3 transmission paths with different groups, which are then configured for the target server a.
[0117] Optionally, in this embodiment, it is assumed that the target server needs to configure 4 transmission paths for a, and the number of path groups of at least two transmission paths is 3. Then, one transmission path is determined from each of the 3 transmission paths, and then any one of the transmission paths is selected from the at least two transmission paths, and one transmission path is determined from any one of the transmission paths, thereby obtaining 3 different groups and 1 duplicate group of transmission paths, which are then configured for the target server a.
[0118] Optionally, in this embodiment, assuming the target server needs to configure 10 transmission paths for pair a, and the number of path groups in at least two sets of transmission paths is 3, then 3 transmission paths are determined from each of the 3 sets of transmission paths. Next, any one set of transmission paths is selected from the at least two sets of transmission paths, and 1 transmission path is determined from that set of paths. This results in 9 transmission paths with different groups and 1 duplicate group, which are then configured for the target server pair a. Therefore, in this embodiment, assuming the number of path groups in at least two sets of transmission paths is n, regardless of the number of transmission paths required for the target server pair a, the transmission paths configured for the target server pair a will have at most n-1 duplicate group transmission paths, where n is an integer greater than or equal to 2.
[0119] It should be noted that, considering the large amount of data, the use of the same switch is usually unavoidable. In order to further reduce the negative impact of the same switch, the difference in the number of transmission paths obtained in each group of transmission paths is limited to less than a preset threshold.
[0120] To illustrate further, we can optionally assume that the preset threshold is 2, and that there are groups 1, 2, and 3. If one transmission path has been determined in group 1 and one transmission path has been determined in group 2, then the number of transmission paths determined in group 3 will not exceed 2. That is, the difference between the number of transmission paths determined in group 3 and the number of transmission paths determined in group 1 or 2 is less than 2.
[0121] Through the embodiments provided in this application, when the number of paths in at least two transmission paths is less than the number of path groups in at least two transmission paths, transmission paths with different numbers of paths are determined from at least two transmission paths; or, when the number of paths in at least two transmission paths is equal to the number of path groups in at least two transmission paths, the same number of transmission paths are obtained from each group of transmission paths in at least two transmission paths; or, when the number of paths in at least two transmission paths is greater than the number of path groups in at least two transmission paths, the same or different number of transmission paths are obtained from each group of transmission paths, wherein the difference in the number of transmission paths obtained from each group of transmission paths is less than a preset threshold, thereby achieving the purpose of reducing the negative impact caused by the same switch, and thus realizing the technical effect of improving data transmission efficiency.
[0122] As an optional approach, in response to a data transmission request triggered by a target server pair, at least two transmission paths are configured for the target server pair based on different groups of transmission paths, including:
[0123] In response to a collaborative training request triggered by a target server pair, at least two transmission paths are configured for the target server pair based on transmission paths of different groups, and the target server pair is instructed to transmit training data streams through at least two transmission paths, wherein the collaborative training request is used to request the target server to perform collaborative training through the transmission of training data streams.
[0124] Optionally, in this embodiment, the collaborative training request is used to request at least two servers to perform collaborative AI training through the transmission of training data streams. Furthermore, this embodiment optimizes the above-mentioned issue of reduced average throughput in cross-regional communication to ensure that traffic on aggregated links in cross-rack communication does not experience self-congestion (e.g., ...). Figure 5 This ensures that the effective throughput of collaborative AI training reaches its optimal level, thereby maximizing the efficiency of collaborative AI training.
[0125] As an optional solution, for ease of understanding, the above data transmission method is applied to the scenario of an AI training cluster. Furthermore, in this embodiment, a scheme to improve the throughput of an AI training cluster in a network where the server uses link aggregation access is proposed. This scheme utilizes the ECMP hash function supported by mainstream switches to achieve relative path control for multiple data streams from the same server node pair. Based on this capability and the source port number detection and allocation design on the server side, the load imbalance problem in the link aggregation scenario is eliminated, thereby eliminating the loop throughput bottleneck of the most important set operation (Ring Allreduce) in mainstream AI training cluster operations, maximizing the overall loop throughput. In an AI training cluster where the server uses link aggregation access, this scheme can improve the effective system throughput by 50% to 100% and reduce training time by 30% to 50%.
[0126] Optionally, such as Figure 13 As shown, this embodiment includes a server-side runtime component. The server-side throughput optimization component in this embodiment proxies the transport protocol stack, accepts requests from applications to establish transport connections, creates a connection using the protocol stack after adding component-specific information, and then returns it to the application. This connection, processed by the components in this embodiment, avoids the traffic aggregation problem described above when transmitting data, maximizing the throughput between communication pairs, thereby maximizing the effective throughput of the entire AI training cluster.
[0127] It should be noted that the technical logic of this embodiment is to eliminate self-congestion by appropriately configuring the sending and source port numbers of multiple connections between a server pair, allowing the two Leaf switches on the receiving server to evenly handle the downlink traffic from the Spine switch. For example... Figure 14 As shown, the technical components of this embodiment consist of two parts: path detection ( Figure 14 Steps 2 and 3 in the process) and connection configuration component ( Figure 14 (Steps 4 and 5 in the previous steps). In this embodiment, the component is located between the business application and the transport protocol stack (communication middleware), receives the application's connection creation request, and proxies the application to request the transport protocol stack to create a transport connection.
[0128] To further illustrate, the specific steps are as follows: Figure 15 As shown, after receiving a transmission request from an application, the component in this embodiment first checks if there is a destination IP address in its local connection configuration information. If so, it extracts the configuration information and uses it to request the creation of a connection from the protocol stack. If not, it initiates a new path probe and stores the probe results in the connection configuration information component for subsequent connection configuration. After obtaining a connection created using the configuration information, the component in this embodiment returns the connection to the application for its data transmission.
[0129] Optionally, for path probing in this embodiment, for a given pair of nodes in a TCP connection or RoCEv2 QP, the four tuples (destination port number, source / destination IP, protocol number) in the five-tuple used for route hashing are usually not arbitrarily modifiable—only the source port number can be arbitrarily modified. Therefore, this embodiment chooses to control the source port number to achieve (to a certain extent) control over the data flow path. In fact, path probing only needs to be performed once, and the results can be persistently saved for subsequent use. The goal of this path probing is to detect K source port numbers that will pass through two non-overlapping paths for the current destination IP (K source port numbers are detected for each path). The two non-overlapping paths are defined as (see...). Figure 5 (The bolded lines in case 2):
[0130] Path 0: From source node sending port 0 to destination node receiving port 0
[0131] Path 1: From source node sending port 1 to destination node receiving port 1
[0132] The non-overlapping path can also be equivalently defined as the path from source sending port 0 to destination receiving port 1 and the path from source sending port 1 to destination receiving port 0. For simplicity, we have chosen only one definition. Specifically, as follows... Figure 16 As shown, the detection steps in this embodiment are as follows:
[0133] Step S5-1: The sending end initiates a probe. First, it detects K source port numbers for path 0. Then, it selects these K source port numbers and uses them to create K probe connections, binding these connections to sending port 0 (i.e., data packets from these connections are sent out from sending port 0). For example, if K = 10, it selects source port numbers 1-10 and creates 10 probe connections to sending port 0 using source port numbers 1-10.
[0134] Step S5-2, Receiver Probe Response. After receiving the probe packet, the receiving server first detects which network interface the packet originated from. Then, it uses the source port number of the probe packet as the source port number of the response packet and sends the ingress information to the sender via the response packet (placed in the payload field). Figure 16 (The dotted line in the diagram). In this embodiment, the detection packet does not directly detect which Leaf switch it passed through to reach the receiving end, but rather determines which Leaf switch it passed through by detecting which network port the detection packet was received from.
[0135] Step S5-3: The sending end processes the probe results. After receiving the response packet, the server extracts the source port number (sport) and the ingress information (ingress) of the payload field. If the ingress meets the conditions (e.g., if the current probe is for path 0, ingress=0 is considered to meet the conditions), the port is added to the connection configuration information. The data structure of the connection configuration information entries is as follows:
[0136] [dip0-path0:sport0_1,sport0_2,sport0_3,...,sport0_N]
[0137] [dip0-path1:sport1_1,sport1_2,sport1_3,...,sport1_N]
[0138] Here, `sport0_1` represents the first source port number of path 0. If the number of source ports for path 0 is less than K, the next batch of K source ports (e.g., 11-20) is used for probing until the number of source ports for path 0 reaches K. If probing of path 0 is complete, the next batch of source ports (e.g., 21-30) is used for probing of path 1 until the number of source ports for path 1 also reaches K. Note that the source port numbers of path 0 and path 1 cannot overlap, otherwise connection creation will fail (the two connection tuples are exactly the same). The number of connections used for inter-server communication in an AI training cluster is usually small, generally two connections using paths 0 and 1 respectively, so K does not need to be configured very large.
[0139] Furthermore, this embodiment always requires path probing for connection creation requests to new destination IPs. However, for the same destination IP, probing only needs to be performed before the first request, and subsequent requests can use the previous probing results. In addition, during AI training, the number of other servers that a server needs to communicate with is generally very small (NCCL's Ring Allreduce operation only needs to communicate with two other servers), so the load of path probing is basically negligible.
[0140] Step S5-4, Connection Configuration. Based on the source port number packets detected in the previous steps, the sending server needs to establish at least two connections when sending data to utilize the bandwidth of all links. Simultaneously, when establishing multiple connections to the same destination server, the following must be done: alternately select the two paths (path0 and path1 defined above) to achieve load balancing of the sending network interface, and sequentially select the source port numbers within the same path's source port number group. For example... Figure 17 In both parts a and b, the results are in line with expectations and the objectives of this embodiment. In part a, the source port number of packet 0 guides the data flow through L2, and the source port number of packet 1 guides the data flow through L3; while in part b, the source port number of packet 0 guides the data flow through L3, and the source port number of packet 1 guides the data flow through L2.
[0141] To further illustrate, assuming the destination server IP is dip0, when requesting two connections, the two connections would use [dip0-path0, sport0_1] and [dip0-path1, sport1_1] respectively. When using three connections, the three connections would use [dip0-path0, sport0_1], [dip0-path1, sport1_1], and [dip0-path0, sport0_2] respectively. When using four connections, the four connections would use [dip0-path0, sport0_1], [dip0-path1, sport1_1], [dip0-path0, sport0_2], and [dip0-path1, sport1_2] respectively, and so on.
[0142] Optionally, as the cluster size increases, the number of cross-rack communications also increases. For existing solutions, at least one Spine downlink traffic aggregation will occur. Figure 5The higher the probability of cases 1 and 4, the more likely the average effective throughput of the system will approach 100G. However, through the embodiments provided in this application, since this embodiment effectively controls the connection path and avoids traffic aggregation, the effective throughput of the system can always maintain the highest performance and will not decay as the cluster size increases. Thus, without the dependence on switch functions, it eliminates the self-congestion of cross-rack aggregated link traffic, maximizes the bandwidth utilization of aggregated links, eliminates the bottleneck of traffic loops in aggregated communication, and maximizes the effective throughput of the AI training cluster.
[0143] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0144] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0145] According to another aspect of the embodiments of this application, a data transmission apparatus for implementing the above-described data transmission method is also provided. For example... Figure 18 As shown, the device includes:
[0146] The determining unit 1802 is used to determine a target server pair from multiple servers in a data center network, wherein the data center network also includes multiple target switches, and the target server pair is two servers communicating across regions.
[0147] The first acquisition unit 1804 is used to acquire data path information of the target server during the data stream transmission process, wherein the data path information is used to indicate the target switch through which the data stream passes during the transmission process.
[0148] Configuration unit 1806 is used to configure at least two sets of transmission paths based on data path information, wherein the transmission paths of the same group in the at least two sets of transmission paths are connected to the same target switch, and the transmission paths of different groups in the at least two sets of transmission paths are connected to different target switches.
[0149] The transmission unit 1808 is configured to, in response to a data transmission request triggered by a target server pair, configure at least two transmission paths for the target server pair based on different groups of transmission paths, and instruct the target server pair to transmit the target data stream through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0150] For specific embodiments, please refer to the examples shown in the above data transmission device, which will not be repeated here.
[0151] As an optional solution, the device also includes:
[0152] The second acquisition unit is used to acquire N probe data streams before acquiring the data path information of the target server during the data stream transmission process. Each probe data stream in the multiple probe data streams is set with a different source port number. The different source port numbers are used to indicate different transmission paths of the probe data streams in the data center network. N is an integer greater than 1.
[0153] The control unit is used to control the target server to transmit N probe data streams before acquiring the data path information during the data stream transmission process of the target server;
[0154] The first acquisition unit 1804 includes: a first acquisition module, used to acquire probe data path information during the transmission process of N probe data streams, wherein the probe data path information is used to indicate the target switch through which the probe data streams pass during the transmission process of the N probe data streams.
[0155] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0156] As an optional solution, configuration unit 1806 includes:
[0157] The execution module performs the following steps until at least two transmission paths are obtained:
[0158] Obtain current data path information, configure at least two candidate paths based on the current data path information, and obtain the number of current candidate paths in each of the at least two candidate paths.
[0159] If the number of current candidate paths is less than the preset number, obtain M probe data streams, where the M probe data streams are the next batch of probe data streams from the N probe data streams;
[0160] Control the target server to transmit M probe data streams;
[0161] Obtain the probe data path information during the transmission process of M probe data streams, and use the probe data path information during the transmission process of M probe data streams as the current data path information. Configure at least two sets of candidate paths based on the current data path information.
[0162] If the number of current candidate paths is greater than or equal to the preset number, at least two candidate paths will be determined as at least two transmission paths.
[0163] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0164] As an optional solution, configuration unit 1806 includes:
[0165] The grouping module is used to group the source port numbers set by multiple probe data streams using data path information to obtain at least two groups of source port numbers. Among them, the probe data streams indicated by the source port numbers of the same group in the at least two groups of source port numbers pass through the same target switch in the transmission path of the data center network, and the probe data streams indicated by the source port numbers of the same group in the at least two groups of source port numbers pass through different target switches in the transmission path of the data center network.
[0166] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0167] As an optional solution, configuration unit 1806 includes:
[0168] The second acquisition module is used to acquire the first transmission path through the first switch and the second transmission path through the second switch based on the data path information, wherein the first switch and the second switch are different first target switches among a plurality of first target switches.
[0169] The determination module is used to determine the first transmission path and the second transmission path as transmission paths in different groups.
[0170] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0171] As an optional solution, the first acquisition unit 1804 includes: a third acquisition module, used to acquire entry information associated with the receiving server, wherein the receiving server is the server that acts as the receiving end in the target server pair, and the data path information includes entry information. The entry information is used to instruct the first network port or the second network port on the receiving server to receive the current data stream. The first network port is connected to the first switch, and the second network port is connected to the second switch.
[0172] The second acquisition module includes: a determination submodule, used to determine the transmission path of the current data stream in the data center network as the first transmission path when the ingress information indicates that the first network port receives the current data stream; or, when the ingress information indicates that the second network port receives the current data stream, to determine the transmission path of the current data stream in the data center network as the second transmission path.
[0173] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0174] As an optional solution, the device also includes:
[0175] The fourth acquisition module is used to acquire the source port number set by the current data flow before acquiring the first transmission path of the first switch and the second transmission path of the second switch based on the data path information. The source port number set by the current data flow is used to indicate the transmission path of the current data flow in the data center network.
[0176] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0177] As an optional solution, the transmission unit 1808 includes:
[0178] The first determining module is configured to determine, when the number of paths in at least two transmission paths is less than the number of path groups in at least two sets of transmission paths, transmission paths of different number of paths; or,
[0179] The second determining module is configured to, when the number of paths in at least two transmission paths is equal to the number of path groups in at least two sets of transmission paths, obtain the same number of transmission paths from each set of transmission paths in at least two sets of transmission paths; or,
[0180] The third determining module is used to obtain the same or different number of transmission paths from each group of transmission paths when the number of paths in at least two transmission paths is greater than the number of path groups in at least two groups of transmission paths, wherein the difference in the number of transmission paths obtained from each group of transmission paths is less than a preset threshold.
[0181] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0182] As an optional solution, the transmission unit 1808 includes:
[0183] The transmission module is used to respond to a collaborative training request triggered by a target server pair, configure at least two transmission paths for the target server pair based on transmission paths of different groups, and instruct the target server pair to transmit training data streams through at least two transmission paths, wherein the collaborative training request is used to request the target server to perform collaborative training through the transmission of training data streams.
[0184] For specific implementation examples, please refer to the examples shown in the above data transmission method; these examples will not be repeated here.
[0185] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described data transmission method is also provided. This electronic device may, but is not limited to, […]. Figure 1 The user equipment 102 or server 112 shown in the figure, in this embodiment, is taken as an example of an electronic device, namely user equipment 102. Further, as shown in the figure... Figure 19 As shown, the electronic device includes a memory 1902 and a processor 1904. The memory 1902 stores a computer program, and the processor 1904 is configured to execute the steps in the above-described method embodiments via the computer program.
[0186] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.
[0187] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:
[0188] S1, Identify the target server pair from multiple servers in the data center network, where the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions;
[0189] S2, obtain the data path information of the target server during the data stream transmission process, wherein the data path information is used to indicate the target switch through which the data stream passes during the transmission process;
[0190] S3, configure at least two sets of transmission paths based on data path information, wherein the transmission paths in the same group of at least two sets of transmission paths have the same target switch, and the transmission paths in different groups of at least two sets of transmission paths have different target switches.
[0191] S4, in response to a data transmission request triggered by a target server pair, configure at least two transmission paths for the target server pair based on transmission paths of different groups, and instruct the target server pair to transmit the target data stream through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0192] Alternatively, as those skilled in the art will understand, Figure 19 The structure shown is for illustrative purposes only. Figure 19 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 19 The more or fewer components shown (such as network interfaces, etc.), or having the same Figure 19 The different configurations shown.
[0193] The memory 1902 can be used to store software programs and modules, such as the program instructions / modules corresponding to the data transmission method and apparatus in this embodiment. The processor 1904 executes various functional applications and data processing by running the software programs and modules stored in the memory 1902, thereby realizing the aforementioned data transmission method. The memory 1902 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1902 may further include memory remotely located relative to the processor 1904, and these remote memories can be connected to electronic devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Specifically, the memory 1902 may be used, but is not limited to, to store information such as data path information, data transmission requests, and target data streams. As an example, such as... Figure 19 As shown, the memory 1902 may include, but is not limited to, the determining unit 1802, the first acquiring unit 1804, the configuration unit 1806, and the transmission unit 1808 from the data transmission device. Furthermore, it may include, but is not limited to, other module units from the data transmission device, which will not be elaborated upon in this example.
[0194] Optionally, the aforementioned transmission device 1906 is used to receive or send data via a network. Specific examples of the network may include wired and wireless networks. In one example, the transmission device 1906 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 1906 is a Radio Frequency (RF) module used for wireless communication with the Internet.
[0195] In addition, the aforementioned electronic device also includes: a display 1908 for displaying information such as the data path information, data transmission request, and target data stream; and a connection bus 1910 for connecting the various module components in the aforementioned electronic device.
[0196] In other embodiments, the aforementioned user equipment or server can be a node in a distributed system, wherein the distributed system can be a blockchain system, which is a distributed system formed by connecting multiple nodes through network communication. The nodes can form a peer-to-peer (P2P) network, and computing devices such as servers, user equipment, and other electronic devices can become nodes in the blockchain system by joining this peer-to-peer network.
[0197] According to one aspect of this application, a computer program product is provided, comprising a computer program / instructions containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions provided in embodiments of this application.
[0198] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0199] It should be noted that the computer system of the electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0200] A computer system includes a Central Processing Unit (CPU), which performs various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) or loaded from RAM. ROM also stores various programs and data required for system operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output interfaces (I / O interfaces) are also connected to the bus.
[0201] The following components are connected to the input / output interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard drives; and communication sections including network interface cards such as LAN cards and modems. The communication section performs communication processing via a network such as the Internet. Drives are also connected to the input / output interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.
[0202] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions defined in the system of this application.
[0203] According to one aspect of this application, a computer-readable storage medium is provided, wherein a processor of a computer device reads computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.
[0204] Optionally, in this embodiment, the computer-readable storage medium described above may be configured to store a computer program for performing the following steps:
[0205] S1, Identify the target server pair from multiple servers in the data center network, where the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions;
[0206] S2, obtain the data path information of the target server during the data stream transmission process, wherein the data path information is used to indicate the target switch through which the data stream passes during the transmission process;
[0207] S3, configure at least two sets of transmission paths based on data path information, wherein the transmission paths in the same group of at least two sets of transmission paths have the same target switch, and the transmission paths in different groups of at least two sets of transmission paths have different target switches.
[0208] S4, in response to a data transmission request triggered by a target server pair, configure at least two transmission paths for the target server pair based on transmission paths of different groups, and instruct the target server pair to transmit the target data stream through at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream.
[0209] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware of an electronic device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0210] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0211] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.
[0212] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0213] In the several embodiments provided in this application, it should be understood that the disclosed user equipment can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.
[0214] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0215] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0216] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.
Claims
1. A data transmission method, characterized by, include: A target server pair is identified from multiple servers in a data center network, wherein the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions. Obtain data path information during the transmission of data streams to the target server, wherein the data path information is used to indicate the target switch through which the data stream passes during the transmission process; Based on the data path information, at least two sets of transmission paths are configured, wherein the transmission paths in the same group of the at least two sets of transmission paths pass through the same target switch, and the transmission paths in different groups of the at least two sets of transmission paths pass through different target switches. In response to a data transmission request triggered by the target server pair, at least two transmission paths are configured for the target server pair based on the transmission paths of the different groups, and the target server pair is instructed to transmit the target data stream through the at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream; The configuration of at least two sets of transmission paths based on the data path information includes: obtaining a first transmission path through a first switch and a second transmission path through a second switch based on the data path information, wherein the first switch and the second switch are different target switches among the plurality of target switches; and determining the first transmission path and the second transmission path as the different sets of transmission paths. The step of obtaining data path information during the data stream transmission process of the target server includes: obtaining entry information associated with the receiving server, wherein the receiving server is the server that acts as the receiving end in the target server pair, the data path information includes the entry information, the entry information is used to instruct the first network port or the second network port on the receiving server to receive the current data stream, the first network port is connected to the first switch, and the second network port is connected to the second switch; The step of obtaining the first transmission path of the first switch and the second transmission path of the second switch based on the data path information includes: when the ingress information is used to indicate that the first network port receives the current data stream, using the transmission path of the current data stream in the data center network as the first transmission path; or, when the ingress information is used to indicate that the second network port receives the current data stream, using the transmission path of the current data stream in the data center network as the second transmission path.
2. The method according to claim 1, characterized in that, Before obtaining the data path information during the data stream transmission process of the target server, the method further includes: N probe data streams are acquired, wherein each of the N probe data streams is set with a different source port number, and the different source port number is used to indicate the different transmission paths of the probe data stream in the data center network, where N is an integer greater than 1; Control the target server to transmit the N probe data streams; The step of obtaining data path information during the transmission of data streams by the target server includes: obtaining probe data path information during the transmission of the N probe data streams, wherein the probe data path information is used to indicate the target switch through which the probe data streams pass during the transmission of the N probe data streams.
3. The method of claim 2, wherein, The configuration of at least two transmission paths based on the data path information includes: Perform the following steps until at least two sets of transmission paths are obtained: Obtain current data path information, configure at least two sets of candidate paths based on the current data path information, and obtain the number of current candidate paths in each set of candidate paths in the at least two sets of candidate paths. If the number of current candidate paths is less than a preset number, M probe data streams are obtained, wherein the M probe data streams are the next batch of probe data streams from the N probe data streams; Control the target server to transmit the M probe data streams; Obtain the detection data path information during the transmission process of the M detection data streams, and use the detection data path information during the transmission process of the M detection data streams as the current data path information, and configure the current at least two sets of candidate paths based on the current data path information; If the number of current candidate paths is greater than or equal to the preset number, the current at least two sets of candidate paths will be determined as the at least two sets of transmission paths.
4. The method of claim 2, wherein, The configuration of at least two transmission paths based on the data path information includes: The source port numbers of the N probe data streams are grouped using the data path information to obtain at least two groups of source port numbers. Among the at least two groups of source port numbers, the source port numbers of the same group indicate that the probe data streams pass through the same target switch in the transmission path of the data center network. The source port numbers of different groups of source port numbers indicate that the probe data streams pass through different target switches in the transmission path of the data center network.
5. The method of claim 1, wherein, Before obtaining the first transmission path through the first switch and the second transmission path through the second switch based on the data path information, the method further includes: Obtain the source port number set for the current data stream, wherein the source port number set for the current data stream is used to indicate the transmission path of the current data stream in the data center network.
6. The method according to any one of claims 1 to 5, characterized in that, The transmission paths based on the different groups are configured for the target server with at least two transmission paths, including: If the number of paths in the at least two transmission paths is less than the number of path groups in the at least two sets of transmission paths, then the different groups of transmission paths representing the number of paths are determined from the at least two sets of transmission paths; or, If the number of paths in the at least two transmission paths is equal to the number of path groups in the at least two sets of transmission paths, then the same number of transmission paths are obtained from each set of transmission paths in the at least two sets of transmission paths; or, When the number of paths in the at least two transmission paths is greater than the number of path groups in the at least two transmission paths, the same or different number of transmission paths are obtained from each group of transmission paths, wherein the difference in the number of transmission paths obtained from each group of transmission paths is less than a preset threshold.
7. The method according to any one of claims 1 to 5, characterized in that, In response to a data transmission request triggered by the target server pair, at least two transmission paths are configured for the target server pair based on the transmission paths of the different groups, including: In response to a collaborative training request triggered by the target server pair, at least two transmission paths are configured for the target server pair based on the transmission paths of the different groups, and the target server pair is instructed to transmit training data streams through the at least two transmission paths, wherein the collaborative training request is used to request the target server to perform collaborative training through the transmission of the training data streams.
8. A data transmission apparatus, characterized by comprising: include: A determining unit is used to determine a target server pair from multiple servers in a data center network, wherein the data center network also includes multiple target switches, and the target server pair consists of two servers communicating across regions. The first acquisition unit is used to acquire data path information of the target server during the transmission of data streams, wherein the data path information is used to indicate the target switch through which the data stream passes during the transmission process; A configuration unit is configured to configure at least two sets of transmission paths based on the data path information, wherein the transmission paths in the same group of the at least two sets of transmission paths are connected to the same target switch, and the transmission paths in different groups of the at least two sets of transmission paths are connected to different target switches. A transmission unit is configured to, in response to a data transmission request triggered by the target server pair, configure at least two transmission paths for the target server pair based on the transmission paths of the different groups, and instruct the target server pair to transmit the target data stream through the at least two transmission paths, wherein the data transmission request is used to request the target server to transmit the target data stream; The configuration of at least two sets of transmission paths based on the data path information includes: obtaining a first transmission path through a first switch and a second transmission path through a second switch based on the data path information, wherein the first switch and the second switch are different target switches among the plurality of target switches; and determining the first transmission path and the second transmission path as the different sets of transmission paths. The step of obtaining data path information during the data stream transmission process of the target server includes: obtaining entry information associated with the receiving server, wherein the receiving server is the server that acts as the receiving end in the target server pair, the data path information includes the entry information, the entry information is used to instruct the first network port or the second network port on the receiving server to receive the current data stream, the first network port is connected to the first switch, and the second network port is connected to the second switch; The step of obtaining the first transmission path of the first switch and the second transmission path of the second switch based on the data path information includes: when the ingress information is used to indicate that the first network port receives the current data stream, using the transmission path of the current data stream in the data center network as the first transmission path; or, when the ingress information is used to indicate that the second network port receives the current data stream, using the transmission path of the current data stream in the data center network as the second transmission path.
9. A computer readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program is executed by an electronic device to perform the method of any one of claims 1 to 7.
10. A computer program product comprising computer programs / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.
11. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.
Citation Information
Patent Citations
Multi-path RDMA transmission
CN115941616A
