Intelligent adjusting and optimizing method and device for RDMA network load, related medium and program product

Abstract

The invention discloses an intelligent adjusting and optimizing method and device for RDMA network loads, a related medium and a program product, and the method comprises the steps: monitoring the number of CNP messages transmitted and received by each interface of a network node, and when the number of CNP messages of any flow is greater than a threshold value, starting the RDMA network loads; and in combination with network topology structure information of the RDMA network and flow state information of each link of a network node, selecting a specific forwarding path with relatively small current flow from at least two forwarding paths reachable by a destination address of any flow, and issuing issuing information corresponding to the flow needing to be forwarded to corresponding network node equipment. According to the intelligent adjusting and optimizing method for the RDMA network load, the link congestion condition of the RDMA network can be sensed, and global flow adjustment is carried out, so that exchange resources in the RDMA network are fully utilized, and the throughput of the RDMA network is improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer networks, and in particular relates to an intelligent tuning method, device, related media and program product for RDMA network load. Background Art

[0002] With the rapid development of computer networks, RDMA (Remote Direct Memory Access) technology has gradually become popular in intelligent computing center networks. RDMA technology has the characteristics of high performance and low latency, so it has high requirements for network stability and reliability. In large-scale networks, the performance of RDMA will be affected due to the complexity of network topology and unpredictable traffic.

[0003] To reduce or avoid network congestion and packet loss and improve the transmission performance of RDMA networks, a common solution is to use PFC (Priority Flow Control) and ECN (Explicit Congestion Notification) technologies for flow control. However, both are passive congestion control mechanisms. They alleviate link congestion by reducing the message sending rate (or stopping message sending) at the source of back pressure traffic, thereby ensuring that the network does not lose packets. However, when the network bandwidth is not fully utilized, they actually reduce network performance.

[0004] In addition, another major cause of network congestion is that each network node in the distributed forwarding structure cannot fully perceive the global network information. It only selects traffic through hash routing based on the number of available links from its own perspective. When a network node sends a message to another network node, it cannot perceive the network status of the downstream node network, resulting in traffic congestion downstream. If other nodes in the network still have available links, the network utilization rate is significantly reduced. Summary of the Invention

[0005] To solve the aforementioned technical problems, the present invention provides an intelligent tuning method for RDMA network load, comprising the following steps:

[0006] S1. Deploy an intelligent network management and control platform in the RDMA network;

[0007] S2. Deploy the collection program on the network node device in the RDMA network;

[0008] S3. The intelligent network management and control platform collects or periodically reads the network traffic status information of the RDMA network;

[0009] S4. The collection program collects network traffic status information, obtains the collected information, and actively uploads the collected information to the intelligent network management and control platform, or the intelligent network management and control platform reads the collected information regularly;

[0010] S5. The intelligent network management and control platform monitors the number of CNP messages sent and received by each interface of the network node;

[0011] S6. When the number of CNP messages of any flow is greater than a threshold, combining the network topology information of the RDMA network and the flow status information of each link of the network node, select a specific forwarding path with smaller current flow from at least two forwarding paths that are reachable to the destination address of any flow;

[0012] S7. The intelligent network management and control platform sends the corresponding information of the traffic to be forwarded to the corresponding network node device.

[0013] Furthermore, the network traffic status information includes network topology information of the RDMA network, traffic status information of each link of the network node, CNP message quantity information of each interface of the network node, and traffic quintuple information associated with the CNP message.

[0014] Furthermore, the delivered information includes quintuple information of traffic and specific forwarding path information.

[0015] Furthermore, in step S6, when it is monitored that the number of CNP messages sent and received by at least two interfaces is greater than a threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted after a first delay time.

[0016] Furthermore, in step S6, when it is monitored that the number of CNP messages sent and received by at least two interfaces is greater than the threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted to the second forwarding path with the second smallest current traffic.

[0017] Furthermore, after step S4 and before step S5, the following steps are further included:

[0018] S', ​​the intelligent network management and control platform visualizes the network traffic status information.

[0019] Furthermore, the method further comprises the following steps:

[0020] S8. The network node device adjusts the traffic to be forwarded to a specific forwarding path according to the information sent down, and forwards it.

[0021] The present invention further provides a computer device comprising a memory, a first processor, and a computer program stored in the memory, wherein the computer program is executed by the first processor and the aforementioned intelligent tuning method for RDMA network load is implemented.

[0022] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program is executed by a second processor to implement the aforementioned intelligent tuning method for the RDMA network load.

[0023] The present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a third processor, the computer program implements the aforementioned intelligent tuning method for RDMA network load.

[0024] Compared with the prior art, the intelligent tuning method for RDMA network load provided by the present invention monitors the number of CNP messages sent and received by each interface of each network node, and when the number of CNP messages of any flow is greater than a threshold, combines the network topology information and the flow status information of each link of the network node, and selects a specific forwarding path with smaller current flow to forward the flow from at least two forwarding paths reachable by the destination address of the any flow. The method can sense the link congestion of the RDMA network, and then perform global flow adjustment based on the network topology and the flow status of each link of the network node, that is, actively adjust the forwarding path of the flow in the upstream network node device, and adjust the congested flow of the downstream network node device to the remaining links with idle resources in the RDMA network, so as to fully utilize the switching resources in the RDMA network and improve the throughput of the RDMA network, thereby improving the overall operation and maintenance level of the data center network, reducing the difficulty of operation and maintenance, and reducing manual intervention in the operation and maintenance process. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flow chart of an intelligent RDMA network load tuning method according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] The following is a detailed description of specific embodiments of the present invention. It should be understood that the embodiments of the present invention are not limited to the embodiments shown in the accompanying drawings, and the scope of protection of the present invention is not limited by the specific embodiments. The terms "first," "second," and similar terms used in the present invention do not indicate any order, quantity, or importance, but are only used to distinguish different components. Similarly, terms such as "one," "an," or "the" do not indicate a quantitative limitation, but rather indicate the presence of at least one. Unless otherwise expressly indicated, throughout the specification and claims, the term "comprise" or variations such as "include" or "comprising" will be understood to include the elements or components stated, without excluding other elements or components. Terms such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0027] Unless otherwise defined, the meanings of all technical terms and scientific terms used in the present invention are the same as those commonly understood by ordinary technicians in the field. In addition, the meanings of the technical terms and scientific terms used in the present invention should be interpreted as having the same meanings as the corresponding terms defined in commonly used technical manuals, and should not be interpreted as having idealized or excessive formal meanings, unless explicitly defined in this way in the present invention.

[0028] Figure 1 A flow chart showing an intelligent RDMA network load tuning method according to an embodiment of the present invention is shown. Figure 1 ,The intelligent tuning method for RDMA network load includes the following steps:

[0029] S1. Deploy an intelligent network management and control platform in the RDMA network.

[0030] Specifically, the intelligent network management and control platform is deployed on the management server of the RDMA network management and control layer.

[0031] S2. Deploy the collection program on the network node device in the RDMA network.

[0032] Specifically, the network node device is located at the network layer of the RDMA network, and the network node device is such as a switch device.

[0033] S3. The intelligent network management and control platform collects or periodically reads the network traffic status information of the RDMA network.

[0034] Specifically, the network traffic status information includes network topology information of the RDMA network, traffic status information of each link of the network node, CNP (Congestion Notification Packet) message quantity information of each interface of the network node, and traffic quintuple information associated with the CNP message.

[0035] Furthermore, the traffic quintuple information includes source IP address, destination IP address, source L4 port, destination L4 port, and protocol number.

[0036] S4. The collection program collects network traffic status information, obtains the collected information, and actively uploads the collected information to the intelligent network management and control platform, or the intelligent network management and control platform reads the collected information regularly.

[0037] The intelligent network management and control platform periodically reads the network traffic status information of the RDMA network from the GPU (Graphics Processing Unit) server, which is located at the computing layer of the RDMA network.

[0038] S5. The intelligent network management and control platform monitors the number of CNP messages sent and received by each interface of the network node.

[0039] S6. When the number of CNP messages of any flow is greater than a threshold, based on the network topology information of the RDMA network and the flow status information of each link of the network node, a specific forwarding path with smaller current flow is selected from at least two forwarding paths that are reachable to the destination address of the any flow.

[0040] Specifically, the threshold value may be manually preset, or may be adjusted according to the actual traffic state of the RDMA network, such as traffic congestion state.

[0041] Specifically, when it is monitored that the number of CNP messages received and sent by at least two interfaces is greater than a threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted after a first delay time.

[0042] Furthermore, the first delay time is adjusted according to an actual traffic state of the RDMA network, such as a traffic congestion state.

[0043] Specifically, when it is monitored that the number of CNP messages sent and received by at least two interfaces is greater than the threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted to the second forwarding path with the second smallest current traffic.

[0044] S7. The intelligent network management and control platform sends the corresponding information of the traffic to be forwarded to the corresponding network node device.

[0045] The information delivered includes the five-tuple information of the traffic and the specific forwarding path information.

[0046] Furthermore, after step S4 and before step S5, the following steps are further included:

[0047] S', ​​the intelligent network management and control platform visualizes the network traffic status information.

[0048] Specifically, the visual display can enable users of the RDMA network to more intuitively understand the current network connection relationship, link traffic conditions, the number of CNP messages sent and received, etc.

[0049] Furthermore, the intelligent tuning method for RDMA network load also includes the following steps:

[0050] S8. The network node device adjusts the traffic to be forwarded to a specific forwarding path according to the information sent down, and forwards it.

[0051] In summary, the intelligent tuning method for RDMA network load provided by the present invention monitors the number of CNP messages received and sent by each interface of each network node, and when the number of CNP messages of any flow is greater than a threshold, combines the network topology information and the flow status information of each link of the network node, and selects a specific forwarding path with smaller current flow to forward the flow from at least two forwarding paths reachable by the destination address of the any flow. It can perceive the link congestion of the RDMA network, and then perform global flow adjustment based on the network topology and the flow status of each link of the network node, that is, actively adjust the forwarding path of the flow in the upstream network node device, and adjust the congested flow of the downstream network node device to the remaining links with idle resources in the RDMA network, so as to fully utilize the switching resources in the RDMA network, improve the throughput of the RDMA network, thereby improving the overall operation and maintenance level of the data center network, reducing the difficulty of operation and maintenance, and reducing manual intervention in the operation and maintenance process.

[0052] The present invention also provides a computer device, comprising a memory, a first processor, and a computer program stored in the memory, wherein the computer program implements the aforementioned intelligent tuning method for RDMA network load when executed by the first processor.

[0053] The present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a second processor, implements the aforementioned intelligent tuning method for RDMA network load.

[0054] The present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a third processor, the computer program implements the aforementioned intelligent tuning method for RDMA network load.

[0055] The foregoing descriptions of specific exemplary embodiments of the present invention are for purposes of illustration and description. These descriptions are not intended to limit the invention to the precise forms disclosed, and it is apparent that many modifications and variations are possible in light of the foregoing teachings. The exemplary embodiments have been selected and described for the purpose of explaining the specific principles of the invention and their practical application, thereby enabling those skilled in the art to realize and utilize a variety of exemplary embodiments of the invention and various options and variations. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. An intelligent tuning method for RDMA network load, characterized in that: The following steps are involved: S1. Deploy an intelligent network management and control platform in the RDMA network; S2, deploying a collection program on a network node device in the RDMA network; S3, the intelligent network management and control platform collects or regularly reads the network traffic status information of the RDMA network; S4. The collection program collects the network traffic status information, obtains the collected information, and actively uploads the collected information to the intelligent network management and control platform, or the intelligent network management and control platform reads the collected information regularly; S5. The intelligent network management and control platform monitors the number of CNP messages sent and received by each interface of the network node; S6. When the number of CNP messages of any flow is greater than a threshold, combining the network topology information of the RDMA network and the flow status information of each link of the network node, select a specific forwarding path with smaller current flow from at least two forwarding paths that are reachable to the destination address of the any flow; S7. The intelligent network management and control platform sends the sending information corresponding to the traffic to be forwarded to the corresponding network node device.

2. The intelligent tuning method for RDMA network load according to claim 1, characterized in that: The network traffic status information includes network topology information of the RDMA network, traffic status information of each link of the network node, CNP message quantity information of each interface of the network node, and traffic quintuple information associated with the CNP message.

3. The intelligent tuning method for RDMA network load according to claim 2, characterized in that: The sent information includes quintuple information of traffic and specific forwarding path information.

4. The intelligent tuning method for RDMA network load according to claim 3, characterized in that: In step S6, when it is monitored that the number of CNP messages sent and received by at least two interfaces is greater than a threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted after a first delay time.

5. The intelligent tuning method for RDMA network load according to claim 3, characterized in that: In step S6, when it is monitored that the number of CNP messages sent and received by at least two interfaces is greater than a threshold, the traffic corresponding to the largest number of CNP messages is adjusted to the first forwarding path with the smallest current traffic, and the traffic of the remaining interfaces is adjusted to the second forwarding path with the second smallest current traffic.

6. The intelligent tuning method for RDMA network load according to claim 3 or 4, characterized in that: After step S4 and before step S5, the following steps are also included: S', ​​the intelligent network management and control platform visualizes the network traffic status information.

7. The intelligent tuning method for RDMA network load according to claim 6, characterized in that: The following steps are also included: S8. The network node device adjusts the traffic to be forwarded to the specific forwarding path according to the issued information and forwards it.

8. A computer device comprising a memory, a first processor, and a computer program stored in the memory, wherein the computer program, when executed by the first processor, implements the intelligent RDMA network load tuning method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a second processor, implements the intelligent RDMA network load tuning method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a third processor, the computer program implements the intelligent tuning method for the RDMA network load according to any one of claims 1 to 7.

Images