Gateway-based method and system for improving wide area network traffic transmission

Abstract

The application provides a gateway-based wide-area network traffic transmission method and system, and relates to the technical field of computer network communication.The method comprises the following steps: an RDMA service host sends an RDMA message to a first gateway through lossless network technology inside a first data center; the first gateway performs wide-area network transmission optimization processing on the RDMA message to obtain an optimized RDMA message; the first gateway transmits the optimized RDMA message to a second gateway through a congestion control algorithm; the second gateway performs stripping path tunnel information processing on the optimized RDMA message to obtain an original RDMA message and sends the original RDMA message to a second data center; and the original RDMA message is sent to a peer RDMA service host through lossless network technology inside the second data center.The application can implement congestion control on the wide-area network transmission process, reduce message loss and retransmission, and improve end-to-end transmission efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer network communication technology, and in particular to a method and system for improving wide area network traffic transmission based on a gateway. Background Technology

[0002] RDMA technology enables hosts to directly access remote memory, featuring low latency, high throughput, and low processor usage. It has been widely used in computing, storage, and artificial intelligence operations within data centers. With the increasing demand for cross-data center computing and data interaction, RDMA services are gradually expanding from within data centers to wide area network (WAN) transmission scenarios.

[0003] Existing technologies typically employ network slicing technologies such as FlexE to establish independent transmission channels for RDMA services across data centers. They also ensure latency, jitter, and packet loss performance during service transmission by using dedicated bandwidth resources, thereby meeting the high requirements of RDMA messages for network reliability and determinism.

[0004] However, existing solutions typically require all network devices at multiple levels in the wide area network to support network slicing, resulting in high costs for device modification and deployment. At the same time, it is difficult to dynamically adjust the transmission process of RDMA messages according to the network congestion status, which can easily lead to message loss or retransmission and reduce end-to-end transmission efficiency. Summary of the Invention

[0005] To address the technical issues that existing solutions typically require all network devices at multiple levels in a wide area network to support network slicing, resulting in high costs for equipment modification and deployment; and that it is difficult to dynamically adjust the transmission process of RDMA messages according to network congestion, which can easily lead to message loss or retransmission and reduce end-to-end transmission efficiency.

[0006] The technical solution provided by this invention is as follows: The first aspect of this invention provides a method for improving wide area network traffic transmission based on a gateway, comprising: S1: The RDMA service host sends RDMA messages to the first gateway through the lossless network technology inside the first data center; S2: The first gateway performs WAN transmission optimization processing on the RDMA message to obtain an optimized RDMA message; S3: Using the congestion control algorithm, the first gateway transmits the optimized RDMA message to the second gateway; S4: The second gateway strips the path tunnel information from the optimized RDMA message to obtain the original RDMA message and sends it to the second data center. S5: Using lossless network technology within the second data center, the original RDMA message is sent to the peer RDMA service host.

[0007] A second aspect of the present invention provides a gateway-based system for improving wide area network traffic transmission, comprising: processor; The memory stores computer-readable instructions, which, when executed by a processor, implement the gateway-based method for improving wide area network traffic transmission as described in the first aspect.

[0008] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the gateway-based method for improving wide area network traffic transmission as described in the first aspect.

[0009] The beneficial effects of the technical solution provided by this invention include: In this embodiment of the invention, the RDMA packets are optimized for wide area network transmission via a first gateway, and the path tunnel information in the optimized RDMA packets is stripped away by a second gateway. This enables the RDMA packets to be transmitted across the wide area network via the two gateways, eliminating the need for all network devices at multiple levels in the wide area network to support network slicing, thereby reducing equipment modification and deployment costs. Simultaneously, the first gateway transmits the optimized RDMA packets to the second gateway using a congestion control algorithm, and packet transmission and reception are completed through lossless network technology within the data centers at both ends. This enables congestion control during the wide area network transmission process, reducing packet loss and retransmission, and improving end-to-end transmission efficiency. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a method for improving wide area network traffic transmission based on a gateway, provided as an embodiment of the present invention; Figure 2 This is a timing diagram of RDMA end-to-end proxy transmission interaction provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a Hash-weighted path selection method provided in an embodiment of the present invention.

[0011] Figure 4 This is a schematic diagram of a congestion control process based on a cache queue and CNP backpressure, provided for an embodiment of the present invention.

[0012] Figure 5 This is a schematic diagram of a Hash index mapping after adjusting the weight of a wide area network transmission path, provided as an embodiment of the present invention.

[0013] Figure 6 This is a schematic diagram of the dynamic adjustment process of WAN transmission path weight provided in an embodiment of the present invention.

[0014] Figure 7 This is a schematic diagram of a gateway-based system for improving wide area network traffic transmission, provided as an embodiment of the present invention. Detailed Implementation

[0015] Reference manual attached Figure 1 The diagram illustrates a flowchart of a method for improving wide area network traffic transmission based on a gateway, as provided in an embodiment of the present invention.

[0016] Reference manual attached Figure 2 The diagram illustrates the timing diagram of RDMA end-to-end proxy transmission interaction provided in an embodiment of the present invention.

[0017] It should be noted that this invention employs a proxy transmission mechanism that coordinates the work of the endpoint, gateway, and wide area network. HostA establishes an end-to-end transmission relationship with the second gateway SGW-B through the first gateway SGW-A. During RDMA service interaction, the first gateway identifies, buffers, and proxies the service flow, and returns a pseudo-ACK message to the sending end. Simultaneously, the first and second gateways perform packet forwarding, congestion control, retransmission, and backpressure control. After receiving the optimized RDMA packet, the second gateway restores the original RDMA packet and forwards it to HostB.

[0018] This invention provides a method for improving wide area network traffic transmission based on a gateway, which may include the following steps: S1: The RDMA service host sends RDMA messages to the first gateway through the lossless network technology inside the first data center.

[0019] In this context, an RDMA service host refers to a computing device, such as a server, storage node, or compute node, that possesses remote direct memory access capabilities and can initiate or receive RDMA service data. RDMA service hosts are typically equipped with network interface cards (NICs) that support the RDMA protocol and corresponding software protocol stacks, enabling direct memory data read / write operations between the local host and the peer host with minimal processor resource consumption.

[0020] Among them, lossless network technology refers to network transmission technology that uses mechanisms such as flow control, congestion notification, priority management, and cache queue management to minimize the dropping of packets due to network congestion or cache overflow.

[0021] It should be noted that the lossless network inside the data center is configured using mainstream technologies such as ECN, PFC, and DCBX.

[0022] In this embodiment of the invention, the lossless network within the first data center can be used to provide a low-packet-loss, low-latency, and low-jitter transmission environment for RDMA messages, ensuring that messages sent by the RDMA service host reliably reach the first gateway and reducing message loss and retransmission caused by network congestion, buffer overflow, or priority conflicts. Simultaneously, by using mechanisms such as ECN, PFC, and DCBX to detect congestion, implement flow control for corresponding priority traffic, and coordinate the lossless transmission parameters of network devices, the stability of the transmission segment within the data center can be guaranteed without changing the original communication method of the RDMA service host.

[0023] S2: The first gateway performs WAN transmission optimization processing on the RDMA message to obtain an optimized RDMA message.

[0024] In one possible implementation, S2 specifically includes: S201: The first gateway identifies the service flow association information of the RDMA message.

[0025] In one possible implementation, S201 specifically includes: S2011: The first gateway determines the RDMA service flow to which the RDMA message belongs based on the port number information, five-tuple information, queue pair information, and address pair information of the RDMA message.

[0026] S2012: The first gateway associates and records the RDMA service flow and the corresponding service flow feature information to obtain the service flow association information.

[0027] S202: Determine the WAN transmission requirements corresponding to the RDMA message based on the service flow association information.

[0028] S203: Determine the WAN transmission path between the first gateway and the second gateway based on WAN transmission requirements.

[0029] Reference manual attached Figure 3 The diagram illustrates a Hash-weighted path selection method provided by an embodiment of the present invention.

[0030] It should be noted that the first gateway first calculates a hash value based on the service flow characteristic information corresponding to the RDMA service flow, and then determines the corresponding WAN transmission path based on the calculated hash index. Each hash index is pre-mapped to a path segment list, and different path segment lists correspond to different WAN transmission paths. In the example shown in the figure, a total of eight hash indices are configured, where hash indices 0, 1, and 2 correspond to the first path segment list List1, hash indices 3, 4, and 5 correspond to the second path segment list List2, and hash indices 6 and 7 correspond to the third path segment list List3, thereby achieving load balancing of different RDMA service flows across multiple WAN transmission paths according to a preset weight ratio.

[0031] In one possible implementation, S203 specifically includes: S2031: Obtain network topology information, link latency information, link jitter information, and link bandwidth information in the wide area network.

[0032] S2032: Establish a network-wide resource model based on network topology information, link latency information, link jitter information, and link bandwidth information.

[0033] The network-wide resource model refers to a data model that uniformly describes the available network resources and operational status of the entire wide area network (WAN) based on the network nodes, links, and their connections. By collecting information such as network topology, link bandwidth, remaining bandwidth, link latency, link jitter, packet loss rate, congestion status, device processing capacity, and path availability, the network-wide resource model is established and dynamically updated to reflect the resource occupancy and transmission performance of different WAN transmission paths.

[0034] S2033: Determine the WAN transmission path based on the network-wide resource model.

[0035] It should be noted that multiple WAN transmission paths are established between the first and second gateways, each corresponding to a specific segment list. The first gateway can employ a weighted multipath forwarding method based on hash indexes. It configures an initial load-sharing weight for each segment list based on at least one of the following factors: link bandwidth, link latency, link jitter, congestion status, and service transmission requirements. It then allocates a corresponding number of hash indices according to the load-sharing weights of each segment list. The first gateway calculates a hash value based on the five-tuple information, queue pair information, address pair information, or other service flow characteristics of the RDMA service flow. It determines the corresponding hash index based on the hash value and transmits the corresponding RDMA service flow through the segment list corresponding to that hash index. For the same RDMA service flow, the first gateway uses the same service flow characteristics to calculate the hash value, ensuring that packets within the same RDMA service flow are preferentially transmitted through the same WAN transmission path, thereby reducing the probability of out-of-order packets within the same service flow.

[0036] For example, three WAN transmission paths are established between the first and second gateways, corresponding to the first path segment list List1, the second path segment list List2, and the third path segment list List3, respectively. The initial load sharing weight of the three WAN transmission paths is 3:3:2. The first gateway sets eight initial hash indices based on the initial load sharing weights, where hash indices 0, 1, and 2 correspond to List1, hash indices 3, 4, and 5 correspond to List2, and hash indices 6 and 7 correspond to List3. The first gateway calculates one of the hash index values ​​from 0 to 7 based on the service flow characteristic information of the RDMA service flow, and selects the corresponding path segment list based on the calculated hash index value. Thus, different RDMA service flows are allocated to the WAN transmission paths corresponding to List1, List2, and List3 according to the 3:3:2 weight ratio.

[0037] S204: The first gateway performs path tunnel encapsulation on the RDMA message based on the WAN transmission path to obtain an optimized RDMA message.

[0038] In one possible implementation, S204 specifically includes: S2041: The first gateway performs buffering processing on RDMA messages.

[0039] S2042: Based on the cache processing results, the service flow characteristic information of the RDMA message, the cache status, and the WAN transmission path status, determine whether the proxy confirmation conditions are met through the first gateway.

[0040] S2043: When the proxy confirmation condition is met, the first gateway returns a pseudo ACK message to the RDMA service host and performs path tunnel encapsulation processing on the RDMA message to obtain an optimized RDMA message.

[0041] In this embodiment of the invention, the first gateway can accurately identify RDMA service flows based on information such as port number, 5-tuple, queue pairs, and address pairs. Combining the bandwidth, latency, jitter, and reliability requirements of the service flow, it selects one or more WAN transmission paths from the network-wide resource model that match its transmission needs, thereby improving the relevance of path selection and network resource utilization. Simultaneously, weighted multi-path forwarding based on hash indexes can reasonably distribute different RDMA service flows across multiple paths and, as far as possible, maintain the same service flow along the same path to reduce the risk of out-of-order packets. Furthermore, the first gateway, through caching and proxy acknowledgment condition judgment, only returns a pseudo-ACK message when the packet has been reliably cached and the transmission path status meets the requirements. This shortens the round-trip latency caused by cross-WAN acknowledgment and improves transmission efficiency. Subsequently, the RDMA packet is transmitted according to the selected path through path tunnel encapsulation, thereby improving the overall throughput and stability of RDMA service transmission across the WAN while considering transmission reliability, latency performance, and path controllability.

[0042] Reference manual attached Figure 4 The diagram illustrates a congestion control process based on a buffer queue and CNP backpressure provided by an embodiment of the present invention.

[0043] It should be noted that the first gateway first classifies and determines the appropriate processing method based on whether the message has experienced a buffer overflow. For RDMA messages that have not experienced a buffer overflow, enqueue processing is performed, and buffer management is implemented based on the buffer queue length and the Pmin and Pmax watermarks. Simultaneously, the LAN-side receiving rate is statistically analyzed using a circular queue, and the current buffer growth is determined in conjunction with the WAN-side sending rate. When the buffer queue length exceeds a preset threshold and the LAN-side receiving rate is greater than the WAN-side sending rate, the first gateway generates a corresponding number of CNP congestion backpressure messages and sends them to the RDMA service host to reduce the sending rate. When the buffer exceeds its maximum capacity, message dropping is performed, thereby preventing continuous buffer growth from causing message loss and improving the stability and end-to-end throughput performance of the wide area network transmission.

[0044] S3: Using a congestion control algorithm, the first gateway transmits the optimized RDMA message to the second gateway.

[0045] Among them, congestion control algorithms refer to control mechanisms that dynamically adjust the message sending rate, sending window, transmission path and retransmission strategy based on information such as network link status, buffer queue length, message transmission rate, latency, packet loss and congestion feedback, in order to avoid aggravating network congestion and improve transmission efficiency.

[0046] It should be noted that if a link failure, multi-party calling, or a mismatch between the LAN and WAN speeds occurs, congestion will occur at some point in the network. In this case, congestion control measures are needed to alleviate and resolve the congestion, thereby restoring the network to normal operation.

[0047] Optionally, the congestion control algorithm specifically includes: flow control algorithm, congestion avoidance algorithm, packet loss detection algorithm, and adaptive algorithm.

[0048] The flow control algorithm takes into account the gateway's buffer occupancy, such as when a packet exceeds the waterline after being enqueued, it will react by issuing congestion notices or CNP backpressure.

[0049] When the congestion avoidance algorithm receives a congestion notice or a CNP backpressure message, it adjusts its packet sending strategy, such as adjusting the rate and window size.

[0050] The packet loss detection algorithm determines whether packet loss has occurred based on recovered messages within a specified time. If packet loss occurs, retransmission should be performed as quickly as possible, using methods such as Go-back-to-N or selective retransmission.

[0051] The adaptive algorithm dynamically adjusts the backpressure frequency, transmission rate, and routing selection based on the real-time network status to respond to changes in network congestion. For example, if the LAN and WAN speeds are mismatched, with the LAN speed being significantly higher than the WAN speed, the backpressure frequency to the LAN will be increased.

[0052] It should be noted that there are multiple paths between SGW gateways. If any path fails, the original load sharing ratio will change, and the data flow will not follow the original path, causing the packet sequence number at both ends to change. This will lead to abnormal transmission of a large number of streams, resulting in a significant decrease in the effective throughput at both ends. Multi-path optimization is designed to avoid mutual impact and only affect the data flow on the original path, without affecting the data flow on other paths, thereby ensuring effective end-to-end throughput.

[0053] In one possible implementation, S3 specifically includes: S301: The first gateway obtains the current buffer queue length, LAN-side receive rate, and WAN-side send rate.

[0054] S302: When the current buffer queue length exceeds the queue threshold and the LAN side receiving rate is greater than the WAN side sending rate, the first gateway generates a congestion backpressure message.

[0055] Those skilled in the art can set the size of the queue threshold according to actual needs, and this invention does not limit it.

[0056] In one possible implementation, S302 specifically includes: S3021: When the current buffer queue length exceeds the queue threshold and the LAN side receiving rate is greater than the WAN side sending rate, obtain the time interval between the current time and the last time the congestion backpressure message was sent.

[0057] S3022: When the time interval is greater than the preset backpressure interval, determine the number of congestion backpressure messages to be sent at the current moment based on the LAN side receiving rate and the WAN side sending rate.

[0058] S3023: Based on the number of messages sent, the first gateway generates a congestion backpressure message.

[0059] Reference manual attached Figure 5 This diagram illustrates a Hash index mapping after adjusting the weight of a wide area network transmission path, as provided in an embodiment of the present invention.

[0060] It should be noted that when congestion or link quality degradation occurs on the WAN transmission path corresponding to the second path segment list List2, the first gateway dynamically adjusts the load-sharing weights of each path segment list based on path quality. In the example shown in the diagram, the load-sharing weight of List2 is reduced, and the number of its corresponding Hash indices is reduced accordingly. Hash index 5, originally mapped to List2, is remapped to List3. Therefore, the adjusted Hash indices 0-2 still correspond to List1, Hash indices 3 and 4 correspond to List2, and Hash indices 5 and 6 correspond to List3. The first gateway selects the corresponding path segment list for forwarding based on the Hash indices calculated from the RDMA service flow, thereby migrating some service flows to the less loaded WAN transmission path, achieving multi-path load balancing and congestion avoidance.

[0061] Reference manual attached Figure 6 This diagram illustrates the dynamic adjustment process of WAN transmission path weights provided in an embodiment of the present invention.

[0062] It should be noted that the first gateway maintains the forwarding relationships corresponding to Hash indices 0-4 unchanged, and modifies the forwarding entry corresponding to Hash index 5 to point to the next forwarding table. When an RDMA service flow hits Hash index 5, the first gateway recalculates the Hash value of the service flow and reselects the corresponding WAN transmission path based on the weighted forwarding table, so that the service flow is reassigned to the corresponding path in List1, List2, or List3, while service flows that do not hit the Hash index to be adjusted retain their original forwarding paths.

[0063] S303: The first gateway sends a congestion backpressure message to the RDMA service host and transmits the optimized RDMA message to the second gateway through the congestion control algorithm.

[0064] It should be noted that during the optimization of RDMA message transmission, the first gateway or service management module continuously acquires at least one of the following data for each WAN transmission path: link latency, link jitter, remaining bandwidth, packet loss rate, congestion status, and service traffic status. Based on the acquired path status, it determines whether the quality of each WAN transmission path has deteriorated. When a target WAN transmission path can no longer carry service traffic according to its original load-bearing weight due to uneven distribution of service flow size, burst traffic, link congestion, increased transmission latency, or decreased link quality, the first gateway reduces the load-bearing weight of the target WAN transmission path and correspondingly reduces the number of Hash indices corresponding to the target WAN transmission path, so as to distribute some of the service traffic on the target WAN transmission path to other available WAN transmission paths.

[0065] For example, when congestion or increased transmission latency occurs on the WAN transmission path corresponding to List2, the first gateway reduces the load-sharing weight of List2 from 3 to 2, and correspondingly reduces the number of hash indices corresponding to List2 from 3 to 2. List2 originally corresponded to hash indices 3, 4, and 5. After the weight adjustment, the mapping relationship between hash indices 3 and 4 and List2 is retained, and hash index 5 is identified as the index to be adjusted. The first gateway changes the forwarding entry corresponding to hash index 5 from directly pointing to List2 to pointing to a pointer to the next forwarding table, causing the RDMA service flow that originally hit hash index 5 to enter the next forwarding table, and recalculates the hash index value for the RDMA service flow. The next forwarding table is generated according to the adjusted weight ratio of 3:2:2, and the RDMA service flows that hit the index to be adjusted are redistributed to the WAN transmission paths corresponding to List1, List2, or List3 based on the recalculated hash index value.

[0066] It should be noted that during the aforementioned weight adjustment, the first gateway maintains the original forwarding indices of List1 and List3 and their corresponding RDMA service flow mapping relationships unchanged, and also maintains the mapping relationships of Hash indices 3 and 4 in List2 that were not selected as the indices to be adjusted. Only the RDMA service flows that originally matched Hash index 5 are recalculated and their paths are reassigned. Therefore, service flows on WAN transmission paths whose weights have not been adjusted are not affected by this weight adjustment. Only service flows that were originally transmitted through the paths corresponding to the indices to be adjusted experience path changes. This avoids a large number of RDMA service flows switching paths simultaneously due to the regeneration of all Hash index mapping relationships, reduces packet out-of-order delivery, retransmissions, and mutual influence between service flows caused by path latency differences, and ensures that the adjusted service traffic sharing ratio meets the 3:2:2 weight requirement.

[0067] Furthermore, when the first gateway detects that the link quality of the WAN transmission path corresponding to List2 has recovered to the preset state, the first gateway can restore the load-sharing weight of List2 from 2 to 3, and refresh the forwarding entry corresponding to Hash index 5. The forwarding entry's pointer, which previously pointed to the next forwarding table, is now directly pointing to List2, allowing subsequent RDMA service flows that hit Hash index 5 to be transmitted again through the WAN transmission path corresponding to List2. By adjusting only the forwarding relationship corresponding to the target Hash index during path quality degradation and recovery, the cascading impact of dynamic changes in path weight on other service flows can be reduced, thereby maintaining the stability of the multi-path transmission process and end-to-end effective throughput.

[0068] Optionally, the first gateway and the second gateway are each configured with independent congestion control strategies. The congestion control algorithm of the first gateway is different from that of the second gateway, in order to adapt to different network congestion characteristics and service requirements.

[0069] It should be noted that although this embodiment mainly describes the first gateway implementing congestion control based on buffer queues and rate matching, in actual deployment, the second gateway can also be configured with an independent congestion control mechanism. For example, the second gateway can dynamically adjust its acknowledgment strategy or generate reverse congestion notifications based on its receive buffer occupancy, the link status when forwarding to the second data center, and the rate at which it receives packets from the first gateway. The congestion control algorithms of the first and second gateways can be different: the first gateway can focus on LAN-WAN rate matching and path selection, while the second gateway can focus on receive-end buffer management and retransmission optimization. This independent and heterogeneous congestion control configuration enables the system to adapt to the differences in the network environments at both ends, improving the overall robustness and throughput efficiency of the transmission.

[0070] In this embodiment of the invention, the first gateway can determine in real time whether there is data backlog by combining the buffer queue length, the LAN-side receiving rate, and the WAN-side sending rate. When congestion conditions are met, it generates a corresponding number of congestion backpressure messages according to the rate difference and the preset backpressure interval, enabling the RDMA service host to reduce the sending rate in a timely manner. This achieves rate matching between the LAN and WAN sides, reducing continuous buffer growth, buffer overflow, and packet loss. Simultaneously, through mechanisms such as flow control, congestion avoidance, packet loss detection and retransmission, and adaptive adjustment, the sending rate, sending window, backpressure frequency, and transmission path can be dynamically adjusted according to network conditions, improving the adaptability of WAN transmission to link failures, traffic bursts, and congestion changes. Furthermore, during multi-path transmission, only the partial hash index and service flow corresponding to the quality-degraded path are remapped, while the service flow forwarding relationship on other paths remains unchanged. This reduces the risk of large-scale service flow migration, packet out-of-order delivery, and retransmission caused by path weight adjustments, thereby improving the stability, reliability, and end-to-end effective throughput of RDMA packet transmission across the WAN.

[0071] S4: The second gateway processes the optimized RDMA message by stripping the path tunnel information, obtains the original RDMA message, and sends it to the second data center.

[0072] It should be noted that the second gateway strips the path tunnel information from the optimized RDMA message, and performs cache management and rate adaptation according to the congestion control policy of the second gateway to obtain the original RDMA message and send it to the second data center.

[0073] In this embodiment of the invention, the second gateway can decapsulate the received optimized RDMA message according to the path tunnel encapsulation format, remove the tunnel information used for WAN path selection and forwarding, and recover the original RDMA message that can be recognized and forwarded by the internal network of the second data center, thereby realizing the protocol connection between the WAN transmission segment and the lossless transmission segment inside the data center. Simultaneously, the second gateway can also verify, deduplicate, or sort the optimized RDMA message based on message integrity, message sequence, and buffer status to reduce the impact of out-of-order, duplicate, or abnormal messages generated by multipath transmission on subsequent RDMA communication, and accurately send the recovered original RDMA message to the second data center, providing a reliable foundation for subsequent transmission to the peer RDMA service host via the lossless network.

[0074] S5: Using lossless network technology within the second data center, the original RDMA message is sent to the peer RDMA service host.

[0075] In this embodiment of the invention, by utilizing the lossless network within the second data center, the original RDMA packets recovered by the second gateway are reliably transmitted to the peer RDMA service host in a manner with low packet loss, low latency, and low jitter. This reduces packet loss, out-of-order delivery, and retransmissions within the data center due to congestion, buffer overflow, or priority contention. Simultaneously, by coordinating and controlling congestion status and priority traffic through mechanisms such as ECN, PFC, and DCBX, a smooth connection between the WAN transmission segment and the transmission segment within the peer data center can be ensured. This allows the peer RDMA service host to receive and process the original RDMA packets according to the original RDMA communication mechanism, thereby maintaining the reliability, stability, and overall throughput performance of end-to-end RDMA service transmission.

[0076] Reference manual attached Figure 7 The diagram illustrates a schematic of a gateway-based wide area network traffic transmission enhancement system provided by an embodiment of the present invention.

[0077] The present invention also provides a gateway-based WAN traffic transmission enhancement system 20, applied to the above-mentioned gateway-based WAN traffic transmission enhancement method, comprising: Processor 201.

[0078] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, they implement the gateway-based method for improving wide area network traffic transmission as described in the method embodiment.

[0079] The gateway-based WAN traffic transmission enhancement system 20 provided by the present invention can execute the above-described gateway-based WAN traffic transmission enhancement method and achieve the same or similar technical effects. To avoid duplication, the present invention will not elaborate further.

[0080] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the gateway-based method for improving wide area network traffic transmission as described in the method embodiments.

[0081] The present invention provides a computer-readable storage medium that can implement the steps and effects of the gateway-based method for improving wide area network traffic transmission in the above-described method embodiments. To avoid repetition, the present invention will not repeat the details.

[0082] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for improving wide area network traffic transmission based on a gateway, characterized in that, include: S1: The RDMA service host sends RDMA messages to the first gateway through the lossless network technology inside the first data center; S2: The first gateway performs wide area network transmission optimization processing on the RDMA message to obtain an optimized RDMA message; S3: Using a congestion control algorithm, the first gateway transmits the optimized RDMA message to the second gateway; S4: The second gateway performs path tunnel information stripping processing on the optimized RDMA message to obtain the original RDMA message and sends it to the second data center; S5: The original RDMA message is sent to the peer RDMA service host through the lossless network technology inside the second data center.

2. The method for improving WAN traffic transmission based on a gateway according to claim 1, characterized in that: S2 specifically includes: S201: The first gateway identifies the service flow association information of the RDMA message; S202: Determine the WAN transmission requirements corresponding to the RDMA message based on the service flow association information; S203: Determine the wide area network transmission path between the first gateway and the second gateway based on the wide area network transmission requirements; S204: The first gateway performs path tunneling encapsulation on the RDMA message based on the WAN transmission path to obtain the optimized RDMA message.

3. The method for improving WAN traffic transmission based on a gateway according to claim 2, characterized in that: S201 specifically includes: S2011: The first gateway determines the RDMA service flow to which the RDMA message belongs based on the port number information, five-tuple information, queue pair information, and address pair information of the RDMA message; S2012: The first gateway associates and records the RDMA service flow and the service flow feature information corresponding to the RDMA service flow to obtain the service flow association information.

4. The method for improving WAN traffic transmission based on a gateway according to claim 2, characterized in that: S203 specifically includes: S2031: Obtain network topology information, link latency information, link jitter information, and link bandwidth information in a wide area network; S2032: Establish a network-wide resource model based on the network topology information, the link latency information, the link jitter information, and the link bandwidth information; S2033: Determine the WAN transmission path based on the network-wide resource model.

5. The method for improving WAN traffic transmission based on a gateway according to claim 2, characterized in that: S204 specifically includes: S2041: The first gateway performs buffering processing on the RDMA message; S2042: Based on the cache processing result, the service flow characteristic information of the RDMA message, the cache status, and the WAN transmission path status, determine whether the proxy confirmation conditions are met through the first gateway. S2043: When the proxy confirmation condition is met, the first gateway returns a pseudo ACK message to the RDMA service host and performs path tunnel encapsulation processing on the RDMA message to obtain the optimized RDMA message.

6. The method for improving WAN traffic transmission based on a gateway according to claim 1, characterized in that: The congestion control algorithms specifically include: flow control algorithm, congestion avoidance algorithm, packet loss detection algorithm, and adaptive algorithm.

7. The method for improving WAN traffic transmission based on a gateway according to claim 1, characterized in that: S3 specifically includes: S301: The first gateway obtains the current buffer queue length, LAN-side receiving rate, and WAN-side sending rate; S302: When the current cache queue length exceeds the queue threshold and the LAN side receiving rate is greater than the WAN side sending rate, the first gateway generates a congestion backpressure message. S303: The first gateway sends the congestion backpressure message to the RDMA service host, and transmits the optimized RDMA message to the second gateway through the congestion control algorithm.

8. The method for improving WAN traffic transmission based on a gateway according to claim 7, characterized in that: S302 specifically includes: S3021: When the current cache queue length exceeds the queue threshold and the LAN side receiving rate is greater than the WAN side sending rate, obtain the time interval between the current time and the last time the congestion backpressure message was sent; S3022: When the time interval is greater than the preset backpressure interval, determine the number of times the congestion backpressure message is sent at the current time based on the LAN side receiving rate and the WAN side sending rate; S3023: Based on the number of messages sent, the first gateway generates the congestion backpressure message.

9. The method for improving WAN traffic transmission based on a gateway according to claim 1, characterized in that: The first gateway and the second gateway are each configured with independent congestion control policies. The congestion control algorithm of the first gateway is different from that of the second gateway to adapt to different network congestion characteristics and service requirements.

10. A gateway-based system for improving wide area network traffic transmission, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the gateway-based method for improving wide area network traffic transmission as described in any one of claims 1 to 9.

Images