Data center network multi-path congestion control method and device based on RoCC

Abstract

The invention provides a multi-path congestion control method and device for a data center network based on RoCC. The method comprises the following steps: before data transmission, pre-constructing a plurality of parallel transmission paths between a source node and a destination node, and configuring congestion control parameters on each switch, including a maximum queue length, a maximum fair rate, a minimum fair rate and a current queue length; in the data transmission process, each hop of switch on a transmission path transmits traffic to a next hop of switch according to a locally maintained price list, calculates the fair rate of the traffic according to the current queue length, generates a feedback message and sends the feedback message to the previous hop of switch and a corresponding source node; the last-hop switch updates a locally maintained price list according to the feedback message, and the source node dynamically selects a transmission path, flow distribution and a sending rate according to the received feedback message.

Description

Technical Field

[0001] The present invention relates to the field of computer network technology, and in particular to a data center network multipath congestion control method and device based on RoCC. Background Art

[0002] With the rapid development of information technology, data center networks have become the core infrastructure supporting modern applications such as cloud computing, big data, and artificial intelligence. Data center networks are growing in scale and complexity. A typical data center may contain tens of thousands of servers interconnected by complex network structures to support distributed applications and massive data transmission. However, as network scale expands and application demands increase, traditional single-path transmission methods face increasing challenges in high-load environments.

[0003] Traditional single-path transmission protocols, such as TCP (Transmission Control Protocol), utilize only a single path in the network for data transmission. This approach works effectively under light network loads, but under high loads, congestion on a single path can lead to packet loss, increased transmission latency, and inefficient use of network resources. Especially in modern data center networks, with the large number of concurrent data flows, single-path transmission can easily cause overload on certain paths, thereby affecting overall network performance. Summary of the Invention

[0004] To solve the above problems, the present invention provides a multi-path congestion control method and device for a data center network based on RoCC. By integrating the RoCC congestion control algorithm and the feedback-based multi-path routing algorithm, the resource utilization of the data center network is improved and the transmission delay is reduced.

[0005] In a first aspect, the present invention provides a data center network multipath congestion control method based on RoCC, comprising:

[0006] Before data transmission, multiple parallel transmission paths are pre-built between the source node and the destination node, and congestion control parameters are configured on each switch; the congestion control parameters include maximum queue length, maximum fair rate, minimum fair rate, and current queue length;

[0007] During data transmission, each hop switch on the transmission path transmits traffic to the next hop switch according to a locally maintained price table, calculates the fair rate of the traffic based on the current queue length, and generates and sends a feedback message to the previous hop switch and the corresponding source node, so that the previous hop switch can update the locally maintained price table based on the feedback message, and the source node can dynamically select the transmission path, traffic distribution, and sending rate based on the received feedback message. The price table is used to store the congestion level and delay information of the transmission path for traffic from any port of the switch to any destination node. The feedback message carries the congestion level and delay information of the best path for traffic to reach the destination node, as well as a traffic identifier and the fair rate of the traffic.

[0008] Furthermore, each hop switch on the transmission path transmits traffic to the next hop switch according to the locally maintained price table. Specifically, when each traffic arrives at a switch, the switch calculates the price of any transmission path p according to the following formula based on the collected feedback information, and preferentially selects the path with the lowest price for data transmission;

[0009]

[0010] Among them H p represents the number of hops on path p, ρ h represents the queue occupancy rate of the h-th hop switch port on the path, d h represents the link transmission delay of the hth hop, ω c With ω d are the weight coefficients of congestion and delay respectively.

[0011] Furthermore, the fair rate of traffic calculation includes: a multiplication reduction strategy and a proportional integral strategy;

[0012] The multiplication reduction stage specifically includes: if the current fair rate F is greater than the preset rate threshold, and the current queue length exceeds the first preset length threshold Q max , the current fair rate is reset to the minimum fair rate; if the current fair rate F is greater than the preset rate threshold, and the queue length change trend exceeds the second preset length threshold Q mid , then the current fair rate is halved;

[0013] The proportional-integral calculation stage specifically includes: a PI controller updating the fair rate based on an input signal and control parameters; wherein the input signal includes the current queue length, the changing trend of the queue length, and the deviation of the current queue length from a queue reference value; the control parameters include a proportional coefficient and an integral coefficient, wherein the proportional coefficient is used to control the degree of influence of the changing trend of the queue length on the fair rate, and the integral coefficient is used to control the degree of influence of the deviation of the current queue length from the reference value on the fair rate.

[0014] Furthermore, the formula for updating the fairness rate of the PI controller is: F'=F-α×(Q cur -Q ref )-β×(Q cur -Q old ); where Q cur is the current queue length, Q ref is the queue reference value, Q cur -Q old is the changing trend of queue length, F' and F represent the fair rate after update and the fair rate before update respectively, Q old Indicates the current queue length after the last adjustment process.

[0015] Furthermore, the previous-hop switch updates the locally maintained price list according to the feedback message, specifically including:

[0016] make The previous hop switch s receives feedback from port i. The switch updates its price table as follows:

[0017]

[0018] Among them, Q i is the queue occupancy rate of port i; d i The link delay can be measured by the timestamp carried in the sub-flow feedback packet.

[0019] Furthermore, the method further includes: after receiving the feedback message, the switch automatically adjusts the parameters of the PI controller according to the network congestion situation; the automatic adjustment process includes:

[0020] Initially, define and set the variable level to 2 and ratio to 0;

[0021] In meeting the conditions When level is less than 64, a loop is entered. In each loop, the level value is doubled to obtain a new level. The new ratio is calculated as the new level divided by 2. The parameters α and β are divided by the new ratio to obtain the adjusted α and β.

[0022] Furthermore, the source node dynamically selects a transmission path, traffic allocation, and sending rate based on the received feedback message, specifically including:

[0023] Based on all received feedback, the congestion window size on each transmission path is dynamically adjusted. If a path becomes congested or has increased delay, the congestion window size of that path is reduced, and part of the traffic is reallocated to other, more unobstructed paths.

[0024] In a second aspect, the present invention provides a data center network multipath congestion control device based on RoCC, comprising:

[0025] A path initialization and parameter configuration module is used to pre-build multiple parallel transmission paths between the source node and the destination node before data transmission, and configure congestion control parameters on each switch; the congestion control parameters include maximum queue length, maximum fair rate, minimum fair rate, and current queue length;

[0026] A feedback and congestion control module is configured to, during data transmission, cause each switch on the transmission path to transmit traffic to the next-hop switch according to a locally maintained price table, calculate a fair rate for the traffic based on the current queue length, generate a feedback message, and send the feedback message to the previous-hop switch and the corresponding source node, so that the previous-hop switch can update the locally maintained price table based on the feedback message, and the source node can dynamically select a transmission path, traffic allocation, and sending rate based on the received feedback message. The price table is configured to store congestion and delay information for the transmission path from any port of the switch to any destination node. The feedback message carries congestion and delay information for the optimal path for the traffic to reach the destination node, as well as a traffic identifier and the fair rate for the traffic.

[0027] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0028] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0029] The beneficial effects of the present invention are:

[0030] By integrating advanced congestion control algorithms and efficient multi-path transmission technologies, this invention significantly improves network resource utilization and transmission efficiency, reduces data transmission delay, and enhances network stability and adaptability. It also optimizes congestion management, reduces reliance on traditional PFC mechanisms, and improves the fairness and real-time performance of data processing, providing data center networks with more efficient, stable, and intelligent traffic management and optimization solutions.

[0031] Through the hop-by-hop feedback mechanism, congestion and delay information of each path in the network can be perceived in real time, thereby making more optimal path selection. This invention supports dynamic path adjustment and can automatically optimize routing strategies according to changes in network status, effectively improving bandwidth utilization and reducing transmission delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of a multi-path congestion control method for a data center network based on RoCC provided in an embodiment of the present invention;

[0033] Figure 2 Flowchart of the RoCC congestion control algorithm provided by an embodiment of the present invention;

[0034] Figure 3 A schematic diagram of the operation of a feedback-based multi-path routing algorithm provided by an embodiment of the present invention;

[0035] Figure 4 A schematic diagram of the structure of a data center network multipath congestion control device based on RoCC provided by an embodiment of the present invention;

[0036] Figure 5 This is a structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] RoCC: Robust Congestion Control for RDMA, a robust congestion control algorithm for RDMA; RDMA is remote direct memory access.

[0039] Combine Figure 1 、 Figure 2 and Figure 3As shown, an embodiment of the present invention provides a data center network multipath congestion control method based on RoCC, comprising the following steps:

[0040] S101: Path initialization and configuration.

[0041] Specifically, before data transmission, MPTCP-RDMA (Multipath TCP with Remote Direct Memory Access) technology is used to pre-build multiple parallel transmission paths between the source and destination nodes, and congestion control parameters are configured on each switch. The congestion control parameters include the maximum queue length, the maximum fair rate, the minimum fair rate, and the current queue length. The congestion control parameters can be maintained in a flow table.

[0042] By using MPTCP-RDMA technology to pre-build multiple parallel transmission paths between the source node and the destination node, it can be ensured that after the data transmission begins, the data sent by the source node can bypass the operating system kernel and directly access the memory.

[0043] It should be noted that after the data transmission process is initiated, a path is selected based on the current state of the network, and the initial data flow is assigned to the selected path. It is understandable that the path with the lowest latency and least congestion will be selected by default.

[0044] S102: During data transmission, each hop switch on the transmission path transmits traffic to the next hop switch according to the locally maintained price table and generates a first feedback message, calculates a fair rate for the traffic according to the current queue length and generates a second feedback message, sends the first feedback message to the previous hop switch, and sends the second feedback message to the corresponding source node, so that the previous hop switch can update the locally maintained price table according to the first feedback message, and the source node can dynamically select a transmission path, traffic distribution, and sending rate according to the received second feedback message.

[0045] The price table is used to store the congestion level cd and delay information l of the transmission path from any port of the switch to any destination node. For example, a single element of the price table can be expressed as (cd, l); the first feedback message carries the congestion level and delay information of the best path for the traffic to reach the destination node; the second feedback message carries the traffic identifier and the fair rate of the traffic.

[0046] Specifically, the switch node sends the second feedback message to the source node via the ICMP protocol. By maintaining a price table on the switch, effective path selection and congestion control are achieved. The price table size is M × K, where M is the number of destination nodes and K is the number of switch ports. Each switch forwards packets to the corresponding port based on MPTCP-RDMA connection information and the congestion level and latency information in the price table, implementing multipath routing and optimizing data transmission paths.

[0047] An embodiment of the present invention provides a multi-path congestion control method for a data center network based on RoCC, which uses the RoCC congestion control algorithm and the feedback-based multi-path routing algorithm to jointly achieve high bandwidth utilization and low round-trip time for the data center network. The RoCC congestion control algorithm uses the current queue length of the switch to calculate the fair rate of each flow in the queue and feeds it back to the source node. After the source node receives the feedback message, the feedback-based multi-path routing algorithm dynamically adjusts the path selection, traffic distribution and sending rate. In addition, a hop-by-hop feedback mechanism is used between switches (i.e., the current switch node sends a feedback message to the previous hop switch and node) to notify the congestion level and delay information within the network, thereby allowing MPTCP-RDMA flows to transparently perform multi-path routing.

[0048] In one embodiment, the fair rate of traffic calculation includes: a multiplication reduction strategy and a proportional integral strategy;

[0049] The multiplication reduction phase specifically includes: if the current fair rate F is greater than the preset rate threshold, and the current queue length Q cur If the first preset length threshold is exceeded, the current fair rate is reset to the minimum fair rate F min If the current fair rate F is greater than the preset rate threshold and the queue length change trend exceeds the second preset length threshold, the current fair rate is halved (i.e., F / 2); As an implementation method, if the maximum fair rate F is set max , then the preset rate threshold is set to F max / 8.

[0050] It is understandable that the first preset length threshold can be directly set to the maximum queue length Q max , the second preset length threshold is set to the queue growth threshold Q mid ;Q max >Q mid In the initial state, data is sent on each transmission path at the maximum sending rate by default. If the network is not congested at this time, the maximum sending rate is the current fair rate; when the network is congested, the sending rate at this time is the current fair rate.

[0051] The proportional integral calculation stage specifically includes: the PI controller updates the fair rate according to the input signal and the control parameter; wherein the input signal includes the current queue length Q cur , the changing trend of queue length Q cur -Q old and the current queue length and queue reference value Q ref The control parameters include a proportional coefficient α and an integral coefficient β, wherein the proportional coefficient is used to control the influence of the change trend of the queue length on the fair rate, and the integral coefficient is used to control the influence of the deviation of the current queue length from the reference value on the fair rate. old Indicates the current queue length after the last adjustment process. In this embodiment, the formula for the PI controller to update the fairness rate is: F'=F-α×(Q cur -Q ref )-β×(Q cur -Q old ). F' and F represent the fair rate after the update and the fair rate before the update, respectively.

[0052] Understandably, Q cur Represents the current congestion level of the network; Q cur -Q old Indicates whether congestion is increasing or decreasing. When it is a negative value, it means that congestion is increasing, otherwise it means that congestion is decreasing. The current queue length is related to Q ref The deviation indicates the stability of the system; Q ref It is the optimal queue length in a stable network state and can be set based on experience; Q max >Q mid >Q ref .

[0053] In summary, in this embodiment, the calculation formula for the fair rate can be expressed as follows:

[0054]

[0055] Specifically, the PI controller continuously adjusts the fair rate until the arrival rate of traffic matches the outflow rate of the congested queue, so the fair rate will stabilize at:

[0056]

[0057] Among them, C l is the bandwidth of the congested path, BW innocent is the total bandwidth used by flows that do not cause congestion, and N is the number of flows that cause congestion.

[0058] In this embodiment, the fair rate adjustment process in the multiplication reduction phase can cope with sudden congestion, and the fair rate adjustment process in the proportional integration phase can maintain the long-term stability of data transmission.

[0059] In one embodiment, after receiving the feedback message, the switch will also automatically adjust the parameters of the PI controller according to the network congestion situation. The automatic adjustment process includes: initially defining and setting the variable level to 2 and ratio to 0. Then, when the condition is met The system enters a loop only if level < 64. In each loop, the level value is doubled, and a new ratio is calculated: the new level divided by 2. Parameters α and β are then scaled based on this new ratio, dividing α and β by the new ratio. Finally, the adjusted α and β are returned, enabling automatic parameter optimization and ensuring system stability and rapid convergence.

[0060] It is worth noting that although the switch stores the information of each port reaching any destination node, in order to reduce communication and storage overhead, in one embodiment, only feedback of the best performing port is provided to the upstream switch.

[0061] For example, for the destination node s, the congestion level and delay of the optimal path from the current switch to the destination node are expressed as (cd [s] ,l [s] ),in:

[0062]

[0063] Among them, D s It is an alternative forwarding port for s.

[0064] The switch periodically sends feedback information to the upstream switch

[0065] When the upstream switch receives these feedback packets, it updates the price table in its local database for future multipath routing decisions. The upstream switch s receives feedback from port i. The switch updates its locally maintained price table as follows:

[0066]

[0067] Among them, Q i is the queue occupancy rate of port i; d i The link delay can be measured by the timestamp carried in the sub-flow feedback packet.

[0068] When each data packet arrives at the switch, the switch calculates the price of each available path based on the collected feedback information. Based on these prices, the system prioritizes the path with the lowest price for data transmission, ensuring optimal utilization of network resources and maximizing transmission performance. When multiple sub-flows are transmitted simultaneously, the switch avoids shared bottlenecks by assigning them to different paths (preferring uncongested paths or paths with minimal congestion), thereby maximizing network resource utilization.

[0069] Specifically, the price includes the cumulative congestion and delay of the path. The price calculation formula for path p is as follows:

[0070]

[0071] Among them H p represents the number of hops on the path, ρ h represents the queue occupancy rate of the h-th hop switch port on the path (used to measure the degree of congestion), d h represents the link transmission delay of the hth hop, ω c With ω d They are the weight coefficients of congestion and delay, respectively, which are used to adjust the relative importance of the two in path selection.

[0072] In one embodiment, the source node dynamically adjusts the transmission path, traffic distribution, and sending rate based on the received feedback message, specifically including:

[0073] By integrating all received feedback messages, the congestion window size on each transmission path is dynamically adjusted. This includes: if a path is congested or the delay increases, the congestion window size of the path is reduced, and part of the traffic is reallocated to other, more unobstructed paths to ensure a balanced load on the entire network.

[0074] In this embodiment, throughout the data transmission process, the source node dynamically adjusts path selection and traffic distribution strategies based on real-time network status feedback. By adjusting the congestion window size, the source node changes its own transmission rate, thereby flexibly varying the transmission rate of each path, ensuring efficient use of network resources and minimizing latency and packet loss.

[0075] In one embodiment, when the source node does not receive a feedback message, it rapidly increases the sending rate through a self-boosting mechanism. The self-boosting mechanism means that when the source node does not receive a feedback message, it assumes that the network is not congested and automatically increases the fair rate of data transmission.

[0076] In one embodiment, if the current data transmission task is completed or the network reaches a preset stable state, the entire congestion control process ends. At this time, all relevant parameters in the data transmission process are reset to prepare for the next data transmission task.

[0077] By integrating advanced congestion control algorithms and efficient multi-path transmission technologies, this invention significantly improves network resource utilization and transmission efficiency, reduces data transmission delay, and enhances network stability and adaptability. It also optimizes congestion management, reduces reliance on traditional PFC mechanisms, and improves the fairness and real-time performance of data processing, providing data center networks with more efficient, stable, and intelligent traffic management and optimization solutions.

[0078] Through the hop-by-hop feedback mechanism, congestion and delay information of each path in the network can be perceived in real time, thereby making more optimal path selection. This invention supports dynamic path adjustment and can automatically optimize routing strategies according to changes in network status, effectively improving bandwidth utilization and reducing transmission delays.

[0079] Based on the same inventive concept, Figure 4 As shown, an embodiment of the present invention further provides a data center network multipath congestion control device based on RoCC, including a path initialization and parameter configuration module and a feedback and congestion control module.

[0080] The path initialization and parameter configuration module is used to pre-establish multiple parallel transmission paths between the source and destination nodes before data transmission and configure congestion control parameters on each switch; the congestion control parameters include the maximum queue length, maximum fair rate, minimum fair rate, and current queue length. During the data transmission process, the feedback and congestion control module is used to cause each hop switch on the transmission path to transmit traffic to the next hop switch according to a locally maintained price table, calculate the fair rate for the traffic based on the current queue length, generate a feedback message, and send it to the previous hop switch and the corresponding source node, so that the previous hop switch can update the locally maintained price table based on the feedback message, and the source node can dynamically select the transmission path, traffic allocation, and sending rate based on the received feedback message. The price table is used to store the congestion level and delay information of the transmission path from any port of the switch to any destination node. The feedback message carries the congestion level and delay information of the optimal path for traffic to reach the destination node, as well as a traffic identifier and the fair rate for the traffic.

[0081] It should be noted that the RoCC-based data center network multipath congestion control device provided in the embodiment of the present invention is intended to implement the above method. Its specific functions can be referred to the above method embodiment and will not be repeated here.

[0082] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5As shown, the electronic device may include: a processor (processor) 501, a communication interface (Communications Interface) 502, a memory (memory) 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. The processor 501 can call logic instructions in the memory 503 to execute a multi-path congestion control method for a data center network based on RoCC. The method includes: before data transmission, pre-establishing multiple parallel transmission paths between a source node and a destination node, and configuring congestion control parameters on each switch; the congestion control parameters include a maximum queue length, a maximum fair rate, a minimum fair rate, and a current queue length; during data transmission, each hop switch on the transmission path transmits traffic to the next hop switch according to a locally maintained price table, calculates a fair rate for the traffic based on the current queue length, and generates and sends a feedback message to the previous hop switch and the corresponding source node, so that the previous hop switch updates the locally maintained price table based on the feedback message, and the source node dynamically selects a transmission path, traffic allocation, and a sending rate based on the received feedback message; wherein the price table is used to store congestion level and delay information of a transmission path for traffic from any port of the switch to any destination node; and the feedback message carries congestion level and delay information of the best path for traffic to reach the destination node, as well as a traffic identifier and the fair rate of the traffic.

[0083] In addition, when the logic instructions in the above-mentioned memory 503 are implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0084] An embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a RoCC-based data center network multipath congestion control method provided in the above-mentioned method embodiments.

[0085] An embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for controlling multipath congestion in a data center network based on RoCC provided in the above-mentioned method embodiments is implemented.

[0086] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multipath congestion control method for data center networks based on RoCC, characterized in that: include: Before data transmission, multiple parallel transmission paths are pre-built between the source node and the destination node, and congestion control parameters are configured on each switch; the congestion control parameters include maximum queue length, maximum fair rate, minimum fair rate, and current queue length; During data transmission, each switch on the transmission path transmits traffic to the next-hop switch according to a locally maintained price table, calculates the fair rate of traffic based on the current queue length, generates a feedback message, and sends it to the previous-hop switch and the corresponding source node. This allows the previous-hop switch to update the locally maintained price table based on the feedback message, and the source node to dynamically select the transmission path, traffic allocation, and transmission rate based on the received feedback message. The price table is used to store the congestion level and delay information of the transmission path from any port of the switch to any destination node. The feedback message carries the congestion level and delay information of the best path for the traffic to reach the destination node, as well as the traffic identifier and the fair rate of the traffic.

2. A data center network multipath congestion control method based on RoCC according to claim 1, characterized in that: Each switch on the transmission path transmits traffic to the next-hop switch based on a locally maintained price table. Specifically, when each traffic arrives at a switch, the switch calculates the price of any transmission path p based on the collected feedback information according to the following formula, and preferentially selects the path with the lowest price for data transmission. Among them H p represents the number of hops on path p, ρ h represents the queue occupancy rate of the h-th hop switch port on the path, d h represents the link transmission delay of the hth hop, ω c With ω d are the weight coefficients of congestion and delay respectively.

3. The RoCC-based data center network multipath congestion control method according to claim 1, characterized in that: The fair rate of the traffic calculation includes: a multiplication reduction strategy and a proportional integral strategy; The multiplication reduction stage specifically includes: if the current fair rate F is greater than the preset rate threshold, and the current queue length exceeds the first preset length threshold Q max , the current fair rate is reset to the minimum fair rate; if the current fair rate F is greater than the preset rate threshold, and the queue length change trend exceeds the second preset length threshold Q mid , then the current fair rate is halved; The proportional-integral calculation stage specifically includes: a PI controller updating the fair rate based on an input signal and control parameters; wherein the input signal includes the current queue length, the changing trend of the queue length, and the deviation of the current queue length from a queue reference value; the control parameters include a proportional coefficient and an integral coefficient, wherein the proportional coefficient is used to control the degree of influence of the changing trend of the queue length on the fair rate, and the integral coefficient is used to control the degree of influence of the deviation of the current queue length from the reference value on the fair rate.

4. The RoCC-based data center network multipath congestion control method according to claim 3, characterized in that: The formula for updating the fairness rate of the PI controller is: F'=F-α×(Q cur -Q ref )-β×(Q cur -Q old ); where Q cur is the current queue length, Q ref is the queue reference value, Q cur -Q old is the changing trend of queue length, F' and F represent the fair rate after update and the fair rate before update respectively, Q old Indicates the current queue length after the last adjustment process.

5. The RoCC-based data center network multipath congestion control method according to claim 1, characterized in that: The previous-hop switch updates the locally maintained price list according to the feedback message, specifically including: make The previous hop switch s receives feedback from port i. The switch updates its price table as follows: Among them, Q i is the queue occupancy rate of port i; d i The link delay can be measured by the timestamp carried in the sub-flow feedback packet.

6. The RoCC-based data center network multipath congestion control method according to claim 3, characterized in that: The method further includes: after receiving the feedback message, the switch automatically adjusts the parameters of the PI controller according to the network congestion situation; the automatic adjustment process includes: Initially, define and set the variable level to 2 and ratio to 0; In meeting the conditions When level < 64, the loop is entered. In each loop, the level value is doubled to obtain a new level. The new ratio is calculated as the new level divided by 2. The parameters α and β are divided by the new ratio to obtain the adjusted α and β.

7. The RoCC-based data center network multipath congestion control method according to claim 1, characterized in that: The source node dynamically selects the transmission path, traffic allocation, and sending rate based on the received feedback message, specifically including: Based on all received feedback, the congestion window size on each transmission path is dynamically adjusted. If a path becomes congested or has increased delay, the congestion window size of that path is reduced, and part of the traffic is reallocated to other, more unobstructed paths.

8. A data center network multipath congestion control device based on RoCC, characterized in that: include: A path initialization and parameter configuration module is used to pre-build multiple parallel transmission paths between the source node and the destination node before data transmission, and configure congestion control parameters on each switch; the congestion control parameters include maximum queue length, maximum fair rate, minimum fair rate, and current queue length; A feedback and congestion control module is configured to, during data transmission, cause each switch on the transmission path to transmit traffic to the next-hop switch according to a locally maintained price table, calculate a fair rate for the traffic based on the current queue length, generate a feedback message, and send the feedback message to the previous-hop switch and the corresponding source node, so that the previous-hop switch can update the locally maintained price table based on the feedback message, and the source node can dynamically select a transmission path, traffic allocation, and transmission rate based on the received feedback message. The price table is configured to store information about the congestion level and delay of the transmission path from any port of the switch to any destination node. The feedback message carries the congestion level and delay information of the best path for the traffic to reach the destination node, as well as the traffic identifier and the fair rate of the traffic.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images