A method and device for implementing load sharing

Abstract

The application provides a load sharing method and device. The load sharing method comprises configuring an out port group; configuring an in port group; wherein the number of in ports of the in port group is equal to the number of out ports of the out port group; configuring a load sharing value LBN value for each out port of the out port group; setting the LBN value of each out port as the LBN value of a mapped in port. The backbone node is ensured to send through a fixed out port based on the in port of the message; and the message is sent through the fixed out port according to the in port LBN value.

Description

Technical Field

[0001] The present application relates to communication technology, and more specifically to a method and device for implementing load sharing. Background Art

[0002] AI (artificial intelligence) computing networks often encounter complex scenarios with multiple input and output ports. In such scenarios, we hope to evenly distribute traffic with the same destination address from multiple input ports to multiple output ports while maintaining packet order (i.e., forwarding each flow). However, in real-world environments, traffic from multiple input ports often converges to a single output port, while other output ports receive no traffic, leading to unbalanced load balancing and network congestion.

[0003] LBN (Load Balance Number) is a network-level load balancing technology that maps equal-cost routing groups to one or more ingress port groups on network devices. Traffic from multiple ingress ports in each port group is precisely load-balanced to the different egress ports within that port group, improving network throughput.

[0004] However, the order of the next-hop egress ports in an equal-cost routing group is not fixed and can change depending on the network environment. Network devices calculate a hash value based on the LBN value of the packet's ingress port and then send the packet through the downgraded egress port corresponding to the calculated hash value. If the order of the next-hop egress ports in an equal-cost routing group changes, packets of the same service flow arriving at the same ingress port in the ingress port group, assuming no routing failures have occurred in the equal-cost routing group, cannot be forwarded along the original path. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for implementing load sharing, which ensures that backbone nodes send messages through fixed egress ports based on the ingress ports at which messages arrive.

[0006] To achieve the above-mentioned purpose, the present application provides a method for implementing load sharing, which includes configuring an egress port group; configuring an ingress port group; wherein the number of ingress ports in the ingress port group is equal to the number of egress ports in the egress port group; configuring a load sharing value LBN value for each egress port in the egress port group; and setting the LBN value of each egress port to the LBN value of the mapped ingress port.

[0007] To achieve the above-mentioned purpose, the present application also provides a device for implementing load sharing, which device at least includes a processor, a memory, and a switching chip connected by a bus; the processor performs the following operations by running the processor executable instructions in the memory: generating an equal-cost routing group to the target IP address; finding that the output port group contains each next-hop output port of the equal-cost routing group; sorting the output port LBN value of each output port in the output port group, and setting each output port in the output port group as the output port of M next-hop table entries of the equal-cost routing group in turn; wherein M=2 N ; M, N are positive integers and M is greater than the number of output ports.

[0008] The beneficial effect of the present application is that a one-to-one correspondence between the input port and the output port between the input port group and the output port group is achieved; the LBN based on the input port sends the message through a fixed output port, avoiding the message of the same business flow arriving at the same input port of the input port group and being sent through different output ports. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of an embodiment of a method for implementing load balancing provided in this application;

[0010] Figure 2 A schematic diagram of a load-sharing link between backbone nodes in an intelligent computing network according to an embodiment of the present application;

[0011] Figure 3 for Figure 2 Schematic diagram of the egress and ingress port groups configured with load sharing values ​​for the backbone nodes;

[0012] Figure 4 for Figure 2 Schematic diagram of configuring the next hop table entry of the equal-cost multi-path group on the backbone node;

[0013] Figure 5 for Figure 2 Schematic diagram of load balancing implemented by backbone nodes;

[0014] Figure 6 for Figure 2 Schematic diagram of the next hop table entry for the configuration fault route of the backbone node;

[0015] Figures 7A-7B A diagram showing how a backbone node distributes the packet load of a failed route to other next hops.

[0016] Figure 8 A schematic diagram of an embodiment of a device for implementing load sharing provided in this application. DETAILED DESCRIPTION

[0017] The present invention will be described in detail with reference to a plurality of examples shown in the accompanying drawings. In the following detailed description, a number of specific details are provided to provide a comprehensive understanding of the present invention. Known methods, steps, components, and circuits are not described in detail in the examples to avoid obscuring the understanding of the examples.

[0018] Among the terms used, the term "including" means including but not limited to; the term "containing" means including but not limited to; the terms "above," "within," and "below" are inclusive; the terms "greater than" and "less than" are exclusive. The term "based on" means based on at least a portion.

[0019] Figure 1 The flowchart of the embodiment of the method for implementing load sharing provided by the present application is shown; the method includes:

[0020] Step 101, configure the egress port group;

[0021] Step 102: configure an ingress port group and ensure that the number of ingress ports in the ingress port group is equal to the number of egress ports in the egress port group.

[0022] Step 103, configuring a load sharing value LBN value for each egress port in the egress port group;

[0023] Step 104: Set the LBN value of each egress port to the LBN value of the mapped ingress port.

[0024] Figure 1 The beneficial effect of the embodiment is that the backbone node sorts each next-hop outbound port of the equal-cost routing group according to the LBN value of each outbound port in the outbound port group, ensuring that the order of the next-hop outbound ports of the equal-cost routing group does not change, and ensuring that when no routing failure occurs in the equal-cost routing group, the message is forwarded according to the original path.

[0025] Figure 2 Schematic diagram of leaf node load sharing links in the intelligent computing network provided in an embodiment of the present application. Figure 2 In the example, backbone nodes Spine 11-14 and leaf nodes Leaf 1 and Leaf 2 are all Layer 3 switching devices. Spine 11 is connected to Leaf 1 through four links connected to ports porta, port b, port c, and port d. Spine 11 is connected to Leaf 2 through four links connected to ports port w, port x, port y, and port z.

[0026] The way spine12-14 connects leaf1 and leaf2 is the same as spine11; for the convenience of illustration, in the embodiment of the present application, the four physical links connecting spine12, spine13, and spine14 to leaf1 are marked as L1, L3, and L5 respectively; the four physical links connecting spine12, spine13, and spine14 to leaf2 are marked as L2, L4, and L6 respectively.

[0027] This application uses spine11 to illustrate the settings of egress port groups, ingress port groups, identification of egress port groups associated with equal-cost routing groups, next-hop port sorting of equal-cost routing groups, and load balancing forwarding.

[0028] Figure 3 for Figure 2 Schematic diagram of the egress and ingress port groups configured with load sharing values ​​for a backbone node.

[0029] Spine11 sets the outbound ports of outbound port group 31 to include port w, port x, port y, and port z.

[0030] Spine11 sets the LBN values ​​of egress ports port w, port x, port y, and port z to LBN0, LBN1, LBN2, and LBN3, respectively.

[0031] Spine11 sets the LBN values ​​of the ingress ports port a, port b, port c, and port d mapped to egress ports port w, port x, port y, and port z to LBN0, LBN1, LBN2, and LBN3, respectively.

[0032] On Spine 11, egress port group 31 is configured first. Then, ingress port group 32, which contains the same number of ingress ports, is configured.

[0033] The ingress port group 32 and egress port group 31 configured on Spine 11 are used to evenly load-share the upstream traffic from leaf 1 onto multiple equal-cost paths to leaf 2.

[0034] Spine11 can also set up another pair of egress port groups and ingress port groups to evenly load-balance the upstream traffic from leaf2 on equal-cost multipaths to leaf1.

[0035] Figure 4 for Figure 2 Schematic diagram of configuring the next hop table entry of the equal-cost multi-path group on the backbone node;

[0036] Spine11 receives routes to server s2 from the four paths connected to ports w, x, y, and z, and generates the routing table entries shown in Table 1.

[0037] Prefix / Mask Next hop Outlet 11.1.1.11 / 32 11.1.1.1 port z 11.1.1.11 / 32 11.1.1.1 port y 11.1.1.11 / 32 11.1.1.1 port x 11.1.1.11 / 32 11.1.1.1 port w

[0038] Table 1

[0039] According to the routing table entries shown in Table 1, Spine11 reaches the four identical next hops 11.1.1.1 of server s2's IP address 11.1.1.11, namely leaf2.

[0040] The four equal-cost routes of Spine11 form equal-cost route group ECMP 33.

[0041] Spine11 finds the egress port group 31 that contains the four next-hop egress ports of equal-cost routing group ECMP 41: port z, port y, port x, and port w.

[0042] Spine11 sorts the four next-hop outbound ports of equal-cost routing group ECMP 33 into port w, port x, port y, and port z according to the LBN values ​​of each outbound port in outbound port group 31: LBN0, LBN1, LBN2, and LBN3. This way, the order of the next-hop outbound ports of the equal-cost routing group remains unchanged.

[0043] Spine11 records the Layer 3 forwarding entry for IP address 11.1.1.11 / 32 in the Layer 3 forwarding table. The next hop of this Layer 3 forwarding entry is ECMP41.

[0044] Spine11 sets each egress port (port w, port x, port y, and port z) of egress port group 31 in turn as the egress port of the 32 next-hop table entries of equal-cost routing group 41.

[0045] In this application, the number of next hop entries in the equal-cost routing group of Spine11 is M=2 N M and N are positive integers, and the number M of next-hop table entries in the equal-cost routing group of Spine 11 is greater than the number of egress ports in egress port group 31. This embodiment uses a full-size 32 next-hop table entries as an example to illustrate how to ensure that the positions of the egress ports of normal routes remain unchanged when a routing failure occurs.

[0046] Figure 5 for Figure 2 Schematic diagram of load balancing implemented by backbone nodes;

[0047] Spine11 receives packet 501 through ingress port a of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates the hash value as hash 0, modulo LBN 0 of ingress port a and the number of members in egress port group 31 associated with ECMP41 (4). It then selects egress port w with hash 0 from ECMP41's 32 next-hop entries. Spine11 re-encapsulates packet 501 with the Ethernet header, using the MAC address of egress port w as the source MAC address and the MAC address of egress port w's next hop, 11.1.1.1, as the destination MAC address. Spine11 then sends the re-encapsulated packet 501 through egress port w.

[0048] Spine11 receives packet 502 through ingress port b in ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates the hash value as hash 1 based on the LBN value LBN1 of ingress port b and the number of members in egress port group 31 associated with ECMP41, modulo 4. It then selects egress port x with hash 1 from ECMP41's 32 next-hop entries. Spine11 re-encapsulates packet 502 with the Ethernet header, using the MAC address of egress port x as the source MAC address and the MAC address of egress port x's next hop, 11.1.1.1, as the destination MAC address. Spine11 then sends the re-encapsulated packet 502 through egress port x.

[0049] Spine11 receives packet 503 through ingress port c of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates the hash value as hash 2, modulo LBN value LBN2 of ingress port b and the number of members in egress port group 31 associated with ECMP41 (4). It then selects egress port y with hash 2 from ECMP41's 32 next-hop table entries. Spine11 uses the MAC address of egress port y as the source MAC address and the MAC address of egress port y's next hop, 11.1.1.1, as the destination MAC address. It then re-encapsulates the Ethernet header for packet 503 and sends the re-encapsulated packet 503 through egress port y.

[0050] Spine11 receives packet 504 through ingress port d of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates a hash value of 3 based on the LBN value LBN3 of ingress port b and the number of members in egress port group 31 associated with ECMP41, modulo 4. It then selects egress port z with hash 3 from ECMP41's 32 next-hop entries. Spine11 uses the MAC address of egress port z as the source MAC address and the MAC address of egress port z's next hop, 11.1.1.1, as the destination MAC address. It then re-encapsulates the Ethernet header for packet 504 and sends the re-encapsulated packet 504 through egress port z.

[0051] Figure 6 for Figure 2 Schematic diagram of the next hop table entry for the configuration fault route of the backbone node;

[0052] The route from port 2 of Spine11 to leaf2 fails. After the route from Spine11 to server s2 is updated, the results are as shown in Table 2.

[0053] Prefix / Mask Next hop Outlet 11.1.1.11 / 32 11.1.1.1 port z 11.1.1.11 / 32 11.1.1.1 port y 11.1.1.11 / 32 11.1.1.1 port x

[0054] Table 2

[0055] Spine11 updates the equal-cost routing group ECMP 41 to three next hops, 11.1.1.1, based on the routing table entries shown in Table 2. Spine11 finds that the three next hops of ECMP41, port z, port y, and port x, belong to egress port group 31.

[0056] Spine 11 compares the three next-hop egress ports of equal-cost routing group ECMP 41 with the egress port of egress port group 31 and identifies that the failed next-hop egress port of egress port group 31 is port w.

[0057] Spine11 identifies the ingress port port a mapped to the faulty egress port port w, and sets an ACL (Access Control List) entry for the ingress port port a. The matching items are port a and ECMP 41, and the action item is to allow forwarding. The packets received by the ingress port port a are load-balanced through the equal-cost routing group 41. By configuring a forwarding entry for port a, the LBN function from the ingress port port a to the faulty egress port port w is disabled, and the packets received by the ingress port port a are borne by the equal-cost multipath group that reaches the destination IP.

[0058] Figures 7A-7B A diagram showing how a backbone node distributes the packet load of a failed route to other next hops.

[0059] Figure 7A In the example, Spine11 receives packet 701 through ingress port b of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop to be ECMP41. Spine11 calculates the hash value as hash 1, modulo LBN 1 of ingress port b of packet 701 and the number of members in egress port group 31 associated with ECMP41 (4). Spine11 selects egress port x with hash 1 from ECMP41's 32 next-hop entries. Using the MAC address of egress port x as the source MAC address and the MAC address of egress port x's next hop 11.1.1.1 as the destination MAC address, Spine11 re-encapsulates the Ethernet header for packet 701. Spine11 then sends the re-encapsulated packet 701 through egress port x.

[0060] Spine11 receives packet 702 through ingress port c of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates a hash value of hash 2 based on the LBN value LBN2 of ingress port b and the number of members in egress port group 31 associated with ECMP41, modulo 4. It then selects egress port y with hash 2 from ECMP41's 32 next-hop table entries. Spine11 re-encapsulates packet 702 with the Ethernet header, using the MAC address of egress port y as the source MAC address and the MAC address of egress port y's next hop, 11.1.1.1, as the destination MAC address. Spine11 then sends the re-encapsulated packet 702 through egress port y.

[0061] Spine11 receives packet 703 through ingress port d of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop as ECMP41. Spine11 calculates a hash value of 3 based on the LBN value LBN3 of ingress port b and the number of members in egress port group 31 associated with ECMP41, modulo 4. It then selects egress port z with hash 3 from ECMP41's 32 next-hop entries. Spine11 re-encapsulates packet 703 with the Ethernet header, using the MAC address of egress port z as the source MAC address and the MAC address of egress port z's next hop, 11.1.1.1, as the destination MAC address. Spine11 then sends the re-encapsulated packet 703 through egress port z.

[0062] Figure 7B In the example, Spine11 receives packet 704 through ingress port a of ingress port group 32. Based on the destination IP address 11.1.1.11, it finds the next hop to be ECMP41. Spine11 obtains a hash factor from packet 704, calculates a hash value modulo the obtained hash factor and the number of ECMP41's next hops, 32, and selects the egress port corresponding to the hash value from ECMP41's 32 next hop table entries. Spine11 uses the MAC address of the egress port's next hop, 11.1.1.1, as the destination MAC address and re-encapsulates the Ethernet header for packet 704. Spine11 sends the re-encapsulated packet 704 through egress port z. Because Spine11 has replaced ECMP41's failed route's egress port w with the next hops x, y, and z of the equal-cost routes in turn, packet 703 received through ingress port 2 is evenly load-balanced by the equal-cost routes' next hops.

[0063] When the path from spine11's port 2 to leaf2 is restored, spine11's routing table 2 to server s2 is updated to the route shown in Table 1; spine11 follows Figure 4 Set the next hop table entry for ECMP41 as shown; spine11 follows Figure 5 In the manner shown, the traffic sent to the server s2 is evenly load-shared again on the four egress ports w, x, y, and z of the egress port group 31 of the LBN.

[0064] Figure 8This is a schematic diagram of an embodiment of a device for implementing load balancing provided by the present application. Device 80 includes at least a processor, memory, and a switch chip connected via a bus. The processor executes processor-executable instructions in the memory to perform the following operations: generate an equal-cost routing group to a target IP address; find an egress port group that includes each next-hop egress port in the equal-cost routing group; and, based on the LBN value of each egress port in the egress port group, alternately set each egress port in the egress port group as the egress port of M next-hop table entries in the equal-cost routing group; where M = 2N; M and N are positive integers, and M is greater than the number of egress ports.

[0065] The processor further performs the following operations by running the processor executable instructions in the memory: generating an equal-cost routing group to the target IP address; finding an egress port group that includes each next-hop egress port of the equal-cost routing group; and setting each egress port of the egress port group as an egress port of M next-hop table entries of the equal-cost routing group in turn based on the LBN value of each egress port of the egress port group; wherein M=2N; M and N are positive integers and M is greater than the number of egress ports.

[0066] The processor further performs the following operations by running processor executable instructions in the memory: receiving a message through one of the ingress ports of the ingress port group; finding an equal-cost routing group according to the destination IP address of the message; calculating a hash value according to the LBN value of the ingress port at which the message arrives and the number of members of the egress port group; selecting an egress port of a next hop table entry of the equal-cost routing group corresponding to the hash value; re-encapsulating an Ethernet header for the message based on the MAC address of the selected egress port; and sending the re-encapsulated message through the selected egress port.

[0067] The processor further performs the following operations by running the processor executable instructions in the memory: searching for an ingress port with the same LBN value based on the LBN value of the egress port of the next hop of the faulty route of the target IP address; configuring a forwarding table entry for the found ingress port to be load-shared by an equal-cost routing group; and replacing the egress ports of the next hop of the faulty route in the M next-hop table entries with the egress ports of the next hop of the normal route of the equal-cost routing group in turn according to the above-mentioned order.

[0068] The processor further performs the following operations by running processor-executable instructions in the memory: refreshing the equivalent route group based on the next hop of the restored route of the target IP address; finding that the output port group includes each next hop output port of the equivalent route group after the restored route; finding an input port with the same LBN value based on the LBN value of the next hop output port of the restored route; deleting the forwarding table entry of the found input port; and setting each output port of the output port group as the output port of M next hop table entries in turn based on the above sorting.

[0069] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for implementing load sharing, characterized in that: The method comprises, Configure the egress port group; Configuring an ingress port group; wherein the number of ingress ports in the ingress port group is equal to the number of egress ports in the egress port group; Configuring a load sharing value LBN value for each egress port of the egress port group; Setting the LBN value of each egress port to the LBN value of the mapped ingress port; Based on the sorting of the LBN value of each egress port in the egress port group, an egress port is selected in turn from the sorted egress ports to forward the message received from the ingress port.

2. The method according to claim 1, characterized in that The method further comprises, Generate an equal-cost routing group to the target IP address; Finding that the egress port group includes each next-hop egress port of the equal-cost routing group; Based on the LBN value sorting of each egress port in the egress port group, each egress port in the egress port group is set as the egress port of the M next hop table entries of the equal-cost routing group in turn; wherein M=2 N ; M and N are positive integers and M is greater than the number of the output ports.

3. The method according to claim 2, characterized in that The method further comprises, Receiving a message through one of the ingress ports of the ingress port group; Find the equal-cost routing group according to the destination IP address of the message; Calculate a hash value based on the LBN value of the ingress port where the message arrives and the number of members of the egress port group; Selecting an egress port of the next hop table entry of the equal-cost routing group corresponding to the hash value; re-encapsulating an Ethernet header for the message based on the MAC address of the selected egress port; The re-encapsulated message is sent through the selected egress port.

4. The method according to claim 2, characterized in that The method comprises, Search for an inbound port with the same LBN value based on the LBN value of the outbound port of the next hop of the faulty route to the target IP address; Configuring a forwarding entry for the found ingress port that is load-shared by the equal-cost routing group; The egress ports of the next hops of the faulty routes in the M next hop table entries are replaced in turn with the egress ports of the next hops of the normal routes of the equal-cost routing group according to the order.

5. The method according to claim 4, characterized in that The method comprises, Refresh the equal-cost routing group based on the next hop of the restoration route of the target IP address; Finding that the egress port group includes each next-hop egress port of the equal-cost routing group after the routing is restored; Based on the LBN value of the egress port of the next hop of the recovery route, find the ingress port with the same LBN value. Deleting the forwarding table entry of the found inbound port; Based on the sorting, each egress port of the egress port group is set in turn as an egress port of the M next hop table entries.

6. A device for implementing load sharing, characterized in that: The device comprises at least a processor, a memory, and a switching chip connected via a bus; the processor performs the following operations by running processor-executable instructions in the memory; Generate an equal-cost routing group to the target IP address; Finding an egress port group that includes each next-hop egress port of the equal-cost routing group; Based on the LBN value sorting of each egress port in the egress port group, each egress port in the egress port group is set as the egress port of the M next hop table entries of the equal-cost routing group in turn; wherein M=2 N ; M and N are positive integers and M is greater than the number of the output ports.

7. The device according to claim 6, characterized in that The processor further performs the following operations by executing the processor-executable instructions in the memory; Generate an equal-cost routing group to the target IP address; Finding that the egress port group includes each next-hop egress port of the equal-cost routing group; Based on the LBN value sorting of each egress port in the egress port group, each egress port in the egress port group is set as the egress port of the M next hop table entries of the equal-cost routing group in turn; wherein M=2 N ; M and N are positive integers and M is greater than the number of the output ports.

8. The device according to claim 6, characterized in that The processor further performs the following operations by executing the processor-executable instructions in the memory; A packet is received through one of the ingress ports in the ingress port group; Find the equal-cost routing group according to the destination IP address of the message; Calculate a hash value based on the LBN value of the ingress port where the message arrives and the number of members of the egress port group; Selecting an egress port of the next hop table entry of the equal-cost routing group corresponding to the hash value; re-encapsulating an Ethernet header for the message based on the MAC address of the selected egress port; The re-encapsulated message is sent through the selected egress port.

9. The device according to claim 6, characterized in that The processor further performs the following operations by executing the processor-executable instructions in the memory; Search for an inbound port with the same LBN value based on the LBN value of the outbound port of the next hop of the faulty route to the target IP address; Configuring a forwarding entry for the found ingress port that is load-shared by the equal-cost routing group; The egress ports of the next hops of the faulty routes in the M next hop table entries are replaced in turn with the egress ports of the next hops of the normal routes of the equal-cost routing group according to the order.

10. The device according to claim 6, characterized in that The processor further performs the following operations by executing the processor-executable instructions in the memory; Refresh the equal-cost routing group based on the next hop of the restoration route of the target IP address; Finding that the egress port group includes each next-hop egress port of the equal-cost routing group after the routing is restored; Based on the LBN value of the egress port of the next hop of the recovery route, find the ingress port with the same LBN value. Delete the forwarding table entry of the found inbound port; Based on the sorting, each egress port of the egress port group is set in turn as an egress port of the M next hop table entries.

Images