Transmission rate control method, apparatus, device, medium, and product

Abstract

The application discloses a transmission rate control method, device, equipment, medium and product, the method comprises the following steps: performing weighted aggregation according to the average round-trip delay of each path under multi-path transmission in a data center network to obtain a fusion round-trip delay of the data center network; controlling the transmission rate of each path in the data center network according to the relative relationship between the average round-trip delay of each path and the fusion round-trip delay and the delay gradient of the fusion round-trip delay, so as to perform path data distribution. The congestion state of each path can be distinguished according to the average round-trip delay and the fusion round-trip delay, and the network state of the data center network can be determined according to the delay gradient, so as to distinguish local congestion from global congestion, thereby reflecting the overall available bandwidth of the multi-path, avoiding inaccurate scheduling decisions, and causing multi-path bandwidth waste or excessive bandwidth utilization.

Description

Technical Field

[0001] This invention relates to the field of data transmission technology, and in particular to a transmission rate control method, apparatus, device, medium, and product. Background Technology

[0002] With the rapid development of artificial intelligence applications, the scale and complexity of data centers are constantly increasing, and the scale and complexity of training models are also continuously growing, placing higher demands on network bandwidth. In this context, the overall bandwidth utilization of the network is becoming increasingly critical. Multipath transmission protocols, due to their ability to utilize clustered parallel links to improve bandwidth utilization, have become a key technology for meeting the high-throughput requirements of AI / ML (Artificial Intelligence / Machine Learning) training, and rate control methods can determine the overall bandwidth utilization performance of multipath transmission. Current rate control technologies rely on single-path congestion signals (packet loss, queuing delay) or historical data prediction models, which struggle to accurately reflect the overall available bandwidth across multiple paths, leading to inaccurate scheduling decisions and resulting in wasted or overused multipath bandwidth. Summary of the Invention

[0003] This invention provides a transmission rate control method, apparatus, device, medium, and product that can distinguish between local congestion and global congestion, so as to reflect the overall available bandwidth of multiple paths, avoid inaccurate scheduling decisions, and prevent multi-path bandwidth waste or over-utilization.

[0004] To achieve the above objectives, embodiments of the present invention provide a transmission rate control method, comprising: The fused round-trip time of the data center network is obtained by weighted aggregation of the average round-trip time of each path under multi-path transmission in the data center network. Based on the relative relationship between the average round-trip time of each path and the fused round-trip time, as well as the delay gradient of the fused round-trip time, the transmission rate of each path in the data center network is controlled to allocate path data.

[0005] As an improvement to the above scheme, the step of controlling the transmission rate of each path in the data center network based on the relative relationship between the average round-trip time of each path and the fused round-trip time, and the delay gradient of the fused round-trip time, to perform path data allocation, includes: The congestion status of each path is determined based on the relative relationship between the average round-trip time of each path and the fused round-trip time. The delay gradient of the fusion round-trip delay is calculated based on the fusion round-trip delay of the previous cycle and the fusion round-trip delay. The rate difference of the data center network is calculated based on the total transmission rate of all paths in the data center network and the total downlink bandwidth of the network. The transmission rate of each path in the data center network is controlled based on the congestion status of each path, the delay gradient, and the rate difference to allocate path data.

[0006] As an improvement to the above scheme, the step of controlling the transmission rate of each path in the data center network based on the congestion status of each path, the delay gradient, and the rate difference to allocate path data includes: The state determination result of the data center network is determined based on the latency gradient and the rate difference; Based on the congestion status of each path and the status determination result, the transmission rate of each path in the data center network is controlled to allocate path data.

[0007] As an improvement to the above scheme, the step of determining the state determination result of the data center network based on the delay gradient and the rate difference includes: If the delay gradient is less than or equal to zero and the rate difference is less than zero, then the state determination result of the data center network is that the bandwidth is idle. If both the delay gradient and the rate difference are greater than zero, then the state determination result is a network congestion state.

[0008] As an improvement to the above scheme, the step of controlling the transmission rate of each path in the data center network based on the congestion status of each path and the status determination result to allocate path data includes: If the state determination result is that the bandwidth is idle, then the transmission rate of each path in the data center network is increased based on the rate difference, the transmission rate of each path, and the link rate. If the state determination result is a network congestion state, then the transmission rate of each path in the data center network is slowed down based on the rate difference, the transmission rate of each path, and the link rate. The packet allocation ratio for each path is determined based on the congestion status of each path and the new transmission rate, and data is allocated to each path in the data center network according to the packet allocation ratio.

[0009] As an improvement to the above scheme, the step of weighted aggregation based on the average round-trip time of each path under multi-path transmission in the data center network to obtain the fused round-trip time of the data center network includes: When the number of acknowledgment messages received in the current period is equal to the number of data packets in flight, the real-time round-trip latency of each path in the multi-path transmission in the data center network is obtained. The real-time round-trip delay of each path is processed using an exponentially weighted moving average algorithm to obtain the average round-trip delay of each path; The average round-trip time of each path is weighted and aggregated based on the link characteristics and historical fluctuations of each path to obtain the converged round-trip time of the data center network.

[0010] To achieve the above objectives, embodiments of the present invention provide a transmission rate control device, comprising: The round-trip delay calculation module is used to perform weighted aggregation based on the average round-trip delay of each path under multi-path transmission in the data center network to obtain the fused round-trip delay of the data center network. The transmission rate control module is used to control the transmission rate of each path in the data center network based on the relative relationship between the average round-trip time of each path and the fused round-trip time, as well as the delay gradient of the fused round-trip time, so as to allocate path data.

[0011] To achieve the above objectives, embodiments of the present invention provide a transmission rate control device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described transmission rate control method.

[0012] To achieve the above objectives, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the above-described transmission rate control method.

[0013] To achieve the above objectives, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the above-described transmission rate control method.

[0014] Compared with existing technologies, the transmission rate control method, apparatus, device, medium, and product disclosed in this invention obtains the fused round-trip time (RTT) of the data center network by weighted aggregation of the average RTT of each path in multi-path transmission. Based on the relative relationship between the average RTT of each path and the fused RTT, and the delay gradient of the fused RTT, the transmission rate of each path in the data center network is controlled for path data allocation. This allows for differentiation of the congestion state of each path based on the average RTT and the fused RTT, and determination of the network state of the data center network based on the delay gradient, thus distinguishing between local and global congestion. This reflects the overall available bandwidth across multiple paths, avoiding inaccurate scheduling decisions that could lead to wasted or overutilized bandwidth. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a transmission rate control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a transmission rate control device provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a transmission rate control device provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that the terms "comprising" and "specific" in this invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0018] Please see Figure 1 , Figure 1 This is a flowchart illustrating a transmission rate control method provided in an embodiment of the present invention. The transmission rate control method includes: S1, the weighted aggregation of the average round-trip time of each path under multi-path transmission in the data center network is performed to obtain the fused round-trip time of the data center network; S2, based on the relative relationship between the average round-trip time of each path and the fused round-trip time, and the delay gradient of the fused round-trip time, the transmission rate of each path in the data center network is controlled to allocate path data.

[0019] For example, the transmission rate control method described in this embodiment of the invention can be implemented by a rate control server, which is capable of interacting with the target user. The rate control server obtains the average round-trip time (RTT) of each path in the multi-path transmission of the data center network; assigns dynamic weights to each path based on link characteristics and historical fluctuations, and performs weighted aggregation of the average RTT to obtain a fused RTT reflecting the overall congestion status of the data center network; calculates the delay gradient of the fused RTT, and distinguishes between congested and lightly loaded paths based on the relative relationship between the average RTT of each path and the fused RTT; and determines the establishment or emptying status of the network queue (network congestion status or bandwidth idle status) based on the sign of the delay gradient. If the delay gradient is positive, the speed of each path is proportionally reduced according to the ratio of traffic transmission rate to link rate; if the delay gradient is negative or zero, the speed of each path is proportionally increased according to 1 minus the ratio of traffic transmission rate to link rate, thereby controlling the multi-path transmission rate of the data center network. The data packet allocation ratio of each path is determined based on the congestion status of each path and the adjusted new transmission rate, and data is allocated among the multiple paths according to the data packet allocation ratio. This invention provides a fused round-trip delay by weighted aggregation of the average round-trip delays of multiple paths. This integrates the dispersed single-path link states into a global congestion quantification indicator, completely solving the technical problem that traditional single-path rate control relies solely on packet loss and delay signals from a single path, failing to reflect the global congestion status of heterogeneous multi-path congestion in a data center. The fused round-trip delay comprehensively reflects the load distribution characteristics of all parallel transmission paths, avoiding misjudgments of congestion caused by instantaneous fluctuations in a single path. It overcomes the limitations of single-path congestion perception, providing a comprehensive and reliable decision-making benchmark for rate control of multi-path transmission. By comparing the relative relationship between the average round-trip delay of a single path and the fused round-trip delay, it clearly defines local link fluctuations and global converged congestion, breaking away from the one-size-fits-all control model for all congestion phenomena. This reflects the overall available bandwidth of multiple paths, enabling rate control strategies to accurately match congestion types. It avoids the waste of bandwidth resources caused by global speed reduction due to local link fluctuations, and also prevents the problem of misjudging global congestion as local fluctuations and allowing congestion to escalate.

[0020] Specifically, step S1 includes: S11, when it is detected that the number of acknowledgment messages received in the current period is equal to the number of data packets in flight, obtain the real-time round-trip delay of each path under multi-path transmission in the data center network; S12, The exponentially weighted moving average algorithm is used to process the real-time round-trip delay of each path to obtain the average round-trip delay of each path; S13, the average round-trip time of each path is weighted and aggregated according to the link characteristics and historical fluctuations of each path to obtain the converged round-trip time of the data center network.

[0021] For example, using round-trip time (RTT) as the core congestion metric in a multi-path environment, a converged RTT is formed by weighted aggregation of the RTT values ​​of each parallel link, which can reflect the overall network status. The RTT of each path is compared with the converged RTT. Links with higher converged RTTs are considered congestion hotspots, while links with lower converged RTTs are considered lightly loaded channels. The converged RTT comprehensively characterizes the load distribution of multiple paths and can identify the overall congestion trend. The relative relationship between path RTTs and converged RTTs can distinguish between local fluctuations and global incast (convergence congestion), triggering load balancing or rate reduction respectively. For example, taking a data center AI training scenario, assume that there are 3 parallel transmission paths (Path1, Path2, Path3) in the data center network, and the link rate (linerate) of all 3 paths is 100Gbps; the total downlink bandwidth of the network is C=250Gbps; set the initial transmission rate (transmission rate of traffic) of each path sub-stream to the corresponding link rate; set the initial value of the average RTT (round-trip time) to the network base latency of 5μs (number of hops × single link transmission latency); initialize the inflight packet count to 0 and the ACK reception count to 0; set the rate control trigger condition to ACK reception count = inflight packet count; initialize the fused RTT weighting coefficient (allocated according to the link bandwidth ratio, with each weighting coefficient being 1 / 3). The system uses real-time statistics of ACK reception counts in the current period and inflight packet counts reset in the previous period. When the number of received acknowledgment messages in the current period equals the number of inflight packets, the system precisely measures the RTT (Path1=7μs, Path2=6μs, Path3=5μs) of each path using the NIC (Network Interface Card) timestamp. The average RTT of each path in the previous period (all 6μs) is also measured. The exponential weighting coefficient α=0.2. The average RTT of each path is updated using an Exponentially Weighted Moving Average (EWMA) algorithm to eliminate instantaneous network fluctuation errors. A weighting coefficient is assigned to the average RTT of each path based on link characteristics and historical fluctuation data. The average round-trip time of each path is then weighted and aggregated to obtain the fused RTT of the data center network. It is worth noting that the fusion weights can be automatically adjusted based on link characteristics and historical fluctuations to ensure robustness and sensitivity under different traffic patterns.This invention dynamically weights and aggregates the round-trip delays of multiple parallel links into a single global congestion metric. This metric not only concisely reflects the load distribution of each path but also adaptively adjusts the weights under different traffic patterns. Therefore, it can obtain accurate multi-path available bandwidth measurements without deploying complex prediction modules or AI agents. The method is highly scalable and compatible with widely deployed functions such as ECN (Explicit Congestion Notification). Utilizing network interface card timestamps and existing ECN support, it does not require special functions on switches or additional complex algorithms deployed on the NIC. Considering the limitations of hardware resources in data center networks, it is fully adaptable to data center environments with limited hardware resources.

[0022] Specifically, step S2 includes: S21, determine the congestion status of each path based on the relative relationship between the average round-trip time of each path and the fused round-trip time; S22, calculate the delay gradient of the fusion round-trip delay based on the fusion round-trip delay of the previous cycle and the fusion round-trip delay; S23, calculate the rate difference of the data center network based on the total transmission rate of all paths in the data center network and the total downlink bandwidth of the network; S24, the transmission rate of each path in the data center network is controlled according to the congestion status of each path, the delay gradient, and the rate difference, so as to perform path data allocation.

[0023] For example, the average RTT of each path is compared with the converged RTT. If the average RTT > the converged RTT, the corresponding path is marked as a congested hotspot path; if the average RTT < the converged RTT, the corresponding path is marked as a lightly loaded path; if the average RTT ≈ the converged RTT, the corresponding path is marked as a load-balanced path. The converged round-trip time (RTT) of the current period is subtracted from the converged RTT of the previous period to obtain the delay gradient ΔRTT of the converged RTT. The sum of the current transmission rates of all paths in the data center network is subtracted from the total downlink bandwidth of the network to obtain the rate difference of the data center network. The transmission rate of each path in the data center network is controlled according to the congestion status of each path, the delay gradient, and the rate difference to perform path data allocation.

[0024] More specifically, step S24 includes: S241, determine the state determination result of the data center network based on the delay gradient and the rate difference; S242, The transmission rate of each path in the data center network is controlled according to the congestion status of each path and the status determination result, so as to perform path data allocation.

[0025] For example, if ΔRTT ≤ 0 and rate difference < 0, the data center network's state is determined to be in a bandwidth idle state; if ΔRTT > 0 and rate difference > 0, the data center network's state is determined to be in a network congestion state. If multiple paths in the data center network are congestion hotspots and ΔRTT ≤ 0, it indicates that the data center network experiences local fluctuations. If the data center network's state is determined to be in a bandwidth idle state, it indicates that there is unused bandwidth in the network. When allocating unused bandwidth, the increment step of the congestion hotspot path should be smaller than the increment step of the lightly loaded path, and the data packets sent by the congestion hotspot path should be allocated to the lightly loaded path. If the data center network's state is determined to be in a network congestion state, it indicates that the network is overutilized in bandwidth, and the decrement step of the congestion hotspot path should be larger than the decrement step of the lightly loaded path. The embodiments of the present invention enable the total transmission rate of multiple paths to quickly converge to the optimal carrying range of the data center network, maximizing bandwidth utilization while ensuring low network latency and no congestion, and meeting the high throughput and low jitter transmission requirements of high-performance services such as AI training and distributed computing.

[0026] More specifically, step S241 includes: S241, if the delay gradient is less than or equal to zero and the rate difference is less than zero, then the state determination result of the data center network is bandwidth idle state; S2412, if both the delay gradient and the rate difference are greater than zero, then the state determination result is network congestion state.

[0027] For example, if ΔRTT≤0 and rate difference<0, i.e. Δq≤0, the queue is empty, the network has unused bandwidth, and the state of the data center network is determined to be in a bandwidth idle state. If ΔRTT>0 and rate difference>0, i.e. Δq>0, the queue is overflowing, the transmission rate exceeds the link capacity, and the state of the data center network is determined to be in a network congestion state.

[0028] More specifically, step S242 includes: S2421, If ​​the state determination result is a bandwidth idle state, then the transmission rate of each path in the data center network is controlled to increase speed based on the rate difference, the transmission rate of each path and the link rate. S2422, If the state determination result is a network congestion state, then the transmission rate of each path in the data center network is slowed down according to the rate difference, the transmission rate of each path and the link rate. S2423, determine the data packet allocation ratio for each path based on the congestion status of each path and the new transmission rate, and allocate data to each path in the data center network according to the data packet allocation ratio.

[0029] For example, when there is unavoidable congestion of the Incast type, suppose time... The network queue length is So for a period of time The change in the length of the inner queue can be expressed as Additionally, assuming the network is a large switch with a fixed downlink bandwidth... Delay gradient for: , , In the formula, The rate difference in a data center network represents the difference between the rate at which the network receives and sends data packets. The total number of traffic flows; Indicates the first The rate at which each data stream is sent; The sum of the transmission rates of all traffic is the rate at which data packets are sent to the network. The difference between the two is the rate at which the network accumulates or clears data packets. RTT (Random Time Tolerance) allows calculation of whether the network queue is increasing or decreasing, and at what rate, enabling precise adjustments.

[0030] If the latency gradient of the fused RTT is negative or zero, it indicates that there is unused bandwidth in the network, the queue is being emptied, and the rate can be increased. The total remaining usable bandwidth of the network is calculated from the latency gradient of the fused RTT. When allocating this remaining usable bandwidth, the step size for increasing the rate of traffic with already high transmission rates should be smaller than that of traffic with low transmission rates. If the latency gradient is positive, it indicates that the network is overutilizing bandwidth, the network is establishing queues, and the rate of traffic should be reduced. In this case, the rate of traffic with already high transmission rates should be reduced more significantly. Specifically, the allocation is determined by the ratio of the transmission rate r to the link rate linerate. ,by This indicates the speed at which the network is building queues, i.e. ,and At this point, the value is greater than 0, the queue is being established, and the flow should be slowed down; for the first... The flow rate decreases at the following formula: , In the formula, Indicates the first The new transmission rate of the traffic (the first) (New transmission rate for the path). This represents the link rate for the corresponding path.

[0031] like ,by This indicates the speed at which the network is emptying the queue. When the value is less than 0, there is idle bandwidth on the network, and the traffic should increase. For the first... The flow rate increases at the following rate: , When there are local fluctuations of the Incast type, the transmission rate of each path in the data center network is not adjusted. A baseline data allocation ratio for each path is calculated based on the new rate, and the ratio is fine-tuned based on the congestion status. The allocation ratio is increased for lightly loaded paths and decreased for congested hotspot paths. Based on the final allocation ratio, data packets in the congestion window are allocated to each parallel path. This embodiment of the invention employs a high-bandwidth-utilization MIMD rate adjustment strategy to address the congestion heterogeneity of multi-path transmission. Through methods such as proportional increase / decrease rates, rapid convergence, and global throughput stability, it fully exploits the potential of parallel links, achieving high throughput and low latency, significantly accelerating the performance of AI / ML training tasks in multi-node ensemble communication scenarios.

[0032] This invention discloses a transmission rate control method that obtains the fused round-trip time (RTT) of the data center network by weighted aggregation of the average RTT of each path in a multi-path transmission system. Based on the relative relationship between the average RTT of each path and the fused RTT, and the delay gradient of the fused RTT, the transmission rate of each path in the data center network is controlled for path data allocation. This method can distinguish the congestion state of each path based on the average RTT and the fused RTT, and determine the network state of the data center network based on the delay gradient, thus differentiating between local and global congestion. This reflects the overall available bandwidth across multiple paths, avoiding inaccurate scheduling decisions that could lead to wasted or overused bandwidth.

[0033] See Figure 2 , Figure 2 This is a schematic diagram of a transmission rate control device 10 provided in an embodiment of the present invention. The transmission rate control device 10 includes: Round-trip delay calculation module 11 is used to perform weighted aggregation based on the average round-trip delay of each path under multi-path transmission in the data center network to obtain the fused round-trip delay of the data center network. The transmission rate control module 12 is used to control the transmission rate of each path in the data center network based on the relative relationship between the average round-trip time of each path and the fused round-trip time, as well as the delay gradient of the fused round-trip time, so as to allocate path data.

[0034] The transmission rate control device 10 provided in this embodiment of the invention can realize all the processes of the transmission rate control method of the above embodiments. The functions and technical effects of each module in the device are the same as the functions and technical effects of the transmission rate control method of the above embodiments, and will not be repeated here.

[0035] See Figure 3 , Figure 3 This is a schematic diagram of a transmission rate control device 20 provided in an embodiment of the present invention. The transmission rate control device 20 of this embodiment includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above-described transmission rate control method embodiment. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module in the above-described transmission rate control device embodiment.

[0036] For example, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the transmission rate control device 20.

[0037] The transmission rate control device 20 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The transmission rate control device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of the transmission rate control device 20 and does not constitute a limitation on the transmission rate control device 20. It may include more or fewer components than shown, or combine certain components, or different components. For example, the transmission rate control device 20 may also include input / output devices, network access devices, buses, etc.

[0038] The processor 21 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 21 is the control center of the transmission rate control device 20, connecting all parts of the transmission rate control device 20 via various interfaces and lines.

[0039] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions of the transmission rate control device 20 by running or executing the computer programs and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0040] If the module integrated into the transmission rate control device 20 is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0041] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0042] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the transmission rate control method as described in the above embodiments.

[0043] Furthermore, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the transmission rate control method described above.

[0044] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A transmission rate control method, characterized in that, include: The fused round-trip time of the data center network is obtained by weighted aggregation of the average round-trip time of each path under multi-path transmission in the data center network. Based on the relative relationship between the average round-trip time of each path and the fused round-trip time, as well as the delay gradient of the fused round-trip time, the transmission rate of each path in the data center network is controlled to allocate path data.

2. The transmission rate control method as described in claim 1, characterized in that, The step of controlling the transmission rate of each path in the data center network based on the relative relationship between the average round-trip time of each path and the fused round-trip time, and the delay gradient of the fused round-trip time, to perform path data allocation, includes: The congestion status of each path is determined based on the relative relationship between the average round-trip time of each path and the fused round-trip time. The delay gradient of the fusion round-trip delay is calculated based on the fusion round-trip delay of the previous cycle and the fusion round-trip delay. The rate difference of the data center network is calculated based on the total transmission rate of all paths in the data center network and the total downlink bandwidth of the network. The transmission rate of each path in the data center network is controlled based on the congestion status of each path, the delay gradient, and the rate difference to allocate path data.

3. The transmission rate control method as described in claim 2, characterized in that, The step of controlling the transmission rate of each path in the data center network based on the congestion status of each path, the delay gradient, and the rate difference to allocate path data includes: The state determination result of the data center network is determined based on the latency gradient and the rate difference; Based on the congestion status of each path and the status determination result, the transmission rate of each path in the data center network is controlled to allocate path data.

4. The transmission rate control method as described in claim 3, characterized in that, The step of determining the state determination result of the data center network based on the delay gradient and the rate difference includes: If the delay gradient is less than or equal to zero and the rate difference is less than zero, then the state determination result of the data center network is that the bandwidth is idle. If both the delay gradient and the rate difference are greater than zero, then the state determination result is a network congestion state.

5. The transmission rate control method as described in claim 3, characterized in that, The step of controlling the transmission rate of each path in the data center network based on the congestion status of each path and the status determination result, in order to allocate path data, includes: If the state determination result is that the bandwidth is idle, then the transmission rate of each path in the data center network is increased based on the rate difference, the transmission rate of each path, and the link rate. If the state determination result is a network congestion state, then the transmission rate of each path in the data center network is slowed down based on the rate difference, the transmission rate of each path, and the link rate. The packet allocation ratio for each path is determined based on the congestion status of each path and the new transmission rate, and data is allocated to each path in the data center network according to the packet allocation ratio.

6. The transmission rate control method as described in claim 1, characterized in that, The step of weighted aggregation based on the average round-trip time of each path in multi-path transmission in the data center network to obtain the fused round-trip time of the data center network includes: When the number of acknowledgment messages received in the current period is equal to the number of data packets in flight, the real-time round-trip latency of each path in the multi-path transmission in the data center network is obtained. The real-time round-trip delay of each path is processed using an exponentially weighted moving average algorithm to obtain the average round-trip delay of each path; The average round-trip time of each path is weighted and aggregated based on the link characteristics and historical fluctuations of each path to obtain the converged round-trip time of the data center network.

7. A transmission rate control device, characterized in that, include: The round-trip delay calculation module is used to perform weighted aggregation based on the average round-trip delay of each path under multi-path transmission in the data center network to obtain the fused round-trip delay of the data center network. The transmission rate control module is used to control the transmission rate of each path in the data center network based on the relative relationship between the average round-trip time of each path and the fused round-trip time, as well as the delay gradient of the fused round-trip time, so as to allocate path data.

8. A transmission rate control device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the transmission rate control method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the transmission rate control method as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product is stored in a storage medium, and the program product is executed by at least one processor to implement the steps of the transmission rate control method as described in any one of claims 1-6.

Images