Multi-host link aggregation for active-active cluster

Patent metadata

Field	Value
Publication number	US-12212486-B2
Application number	US-202318308600-A
Country	US
Kind code	B2
Filing date	Apr 27, 2023
Priority date	Apr 27, 2023
Publication date	Jan 28, 2025
Grant date	Jan 28, 2025

Abstract

Embodiments of multi-host link aggregation are described for active-active cluster implementations. In one or more embodiments, a multi-host LAG bond in the cluster is created with a bond MAC address designated for the bond. Afterwards, a virtual IP address may be created for all hosts within the multi-host LAG bond. Compared with existing active-active cluster implementation, the presented multi-host link aggregation embodiments do not require extra entities, e.g., load balancer or DNS, which could be a failure point or performance bottleneck, thus makes the active-active cluster implementations more efficient and robust. With the Multi-Host LACP bond and virtual IP on all hosts, the cluster may have increased bandwidth to the cluster service and utilize computing capacity in a network switch for improved workload distribution.

First claim

What is claimed is:

1. A processor-implemented method for implementing multi-host link aggregation group (LAG), the method comprising:
given an active-active cluster comprising a plurality of hosts with each host having an agent, performing steps comprising:
assigning an agent on a host among the plurality of hosts as a primary agent;
setting one or more agents on one or more hosts other than the host where the primary agent resides as normal agents;
setting, by the primary agent, a bond Media Access Control (MAC) address for a multi-host link aggregation group (LAG) bond for the plurality of hosts to a MAC address of a local port of the host;
exchanging information between the primary agent and the normal agents; and
creating a service IP (Internet Protocol) address for all hosts in the multi-host LAG bond, the service IP address is a virtual IP address for the bond MAC address.

2. The processor-implemented method of claim 1 wherein the active-active cluster utilizes Link Aggregation Control Protocol (LACP) bond frame distribution mechanisms to perform load balancing among the plurality of hosts.

3. The processor-implemented method of claim 1 wherein each host has a single link or multiple links to a network switch.

4. The processor-implemented method of claim 1 wherein the step of exchanging information between the primary agent and the normal agents comprises steps of:
initiating, by the primary agent, a local bond member table for the primary agent comprising one or more networking information;
initiating, by the primary agent, a remote bond member table for the primary agent, which is an empty table upon initialization;
sending, from the primary agent, a query request comprising the bond MAC address to all normal agents; and
updating, by the primary agent, the remote bond member table for the primary agent based on query responses received from one or more normal agents.

5. The processor-implemented method of claim 4 wherein the step of exchanging information between the primary agent and the normal agents comprises steps of:
initiating, by each of the normal agents, a remote bond member table for each normal agent, which is an empty table upon initialization;
upon receiving the query request from the primary agent, changing, by each of the normal agents, a MAC address of a local link to the bond MAC address;
initiating, by each of the normal agents, a local bond member table for each normal agent comprising one or more networking information; and
sending, from each of the normal agents to the primary agent, a query response comprising a local bond member status.

6. The processor-implemented method of claim 1 wherein the hosts of the plurality of hosts are interconnected by a management network, which causes the multi-host LAG bond in the active-active cluster to be created.

7. The processor-implemented method of claim 1 wherein:
responsive to an Address Resolution Protocol (ARP) request being sent from a client served by the active-active cluster, a single host among the plurality of hosts in the multi-host LAG bond receives and responds to the ARP request.

8. The processor-implemented method of claim 1 further comprising:
performing steps of initializing a Transmission Control Protocol (TCP) session comprising:
receiving, at a port on a network switch coupled to the multi-host LAG bond, a TCP synchronization (SYN) packet from a client by the active-active cluster, the TCP SYN packet comprises a destination MAC address that is the bond MAC address, a source MAC address that is the MAC address of the client, a destination IP address that is the virtual IP, and a source IP address that is the IP address of the client;
forwarding, by the network switch, the TCP SYN packet to a port channel;
distributing, by a port channel aggregator on the network switch, the TCP SYN packet to a link;
receiving, at a host connected to the link, the TCP SYN packet;
sending, by the host connected to the link, a TCP SYN acknowledgement (ACK) packet to the client, the TCP SYN ACK packet comprises a destination MAC address that is the MAC address of the client, a source MAC address that is the bond MAC address, a destination IP address that is the IP address of the client, and a source IP address that is the virtual IP address;
sending, by the client, a TCP ACK packet; and
distributing, by a port channel aggregator on the network switch, the TCP ACK packet to the link.

9. A non-transitory computer-readable medium or media comprising one or more sequences of instructions which, when executed by at least one processor, causes steps to be performed comprising:
assigning an agent, among a plurality of agents with an agent installed on each of a plurality of hosts in an active-active cluster, on a host among the plurality of hosts as a primary agent;
setting one or more agents, among the plurality of agents, other than the primary agent as normal agents;
setting a bond Media Access Control (MAC) address for a multi-host link aggregation group (LAG) bond for the plurality of hosts to a MAC address of a local port of the host where the primary agent resides;
exchanging information between the primary agent and the one or more normal agents; and
creating a service IP (Internet Protocol) address for all hosts in the multi-host LAG bond, the service IP address is a virtual IP address for the bond MAC address.

10. The non-transitory computer-readable medium or media of claim 9 wherein the steps further comprise:
utilizing Link Aggregation Control Protocol (LACP) bond frame distribution mechanisms to perform load balancing among the plurality of hosts.

11. The non-transitory computer-readable medium or media of claim 9 wherein the step of exchanging information between the primary agent and the normal agents comprises steps of:
initiating, by the primary agent, a local bond member table for the primary agent comprising one or more networking information;
initiating, by the primary agent, a remote bond member table for the primary agent, which is an empty table upon initialization;
sending, from the primary agent, a query request comprising the bond MAC address to all normal agents; and
updating, by the primary agent, the remote bond member table for the primary agent based on query responses received from one or more normal agents.

12. The non-transitory computer-readable medium or media of claim 11 wherein the step of exchanging information between the primary agent and the normal agents further comprises steps of:
initiating, by each normal agent, a remote bond member table for the normal agent, which is an empty table upon initialization;
upon receiving the query request from the primary agent, changing, by the normal agent, a MAC address of a local link to the bond MAC address;
initiating, by the normal agent, a local bond member table for the normal agent comprising one or more networking information; and
sending, from the normal agent to the primary agent, a query response comprising a local bond member status.

13. The non-transitory computer-readable medium or media of claim 9 wherein the steps further comprise:
responsive to an Address Resolution Protocol (ARP) request sent from a client served by the active-active cluster, a single host among the plurality of hosts in the multi-host LAG bond receiving and responding to the ARP request.

14. The non-transitory computer-readable medium or media of claim 9 wherein the steps further comprise:
initializing a Transmission Control Protocol (TCP) session using steps comprising:
receiving, at a port on a network switch coupled to the multi-host LAG

Assignees

• Dell Products Lp

Inventors

• Yang Jianfei
• Luo Mingyi
• Pu Weilan
• Chamarthy Srinivasa Raju
• Zheng Lifeng

Classifications

• H04L61/58

  Caching of addresses or names · CPC title

• H04L61/5007

  Internet protocol [IP] addresses · CPC title

• H04L61/2596

  Translation of addresses of the same type other than IP, e.g. translation from MAC to MAC addresses · CPC title

• H04L2101/668

  Internet protocol [IP] address subnets · CPC title

• H04L2101/622

  Layer-2 addresses, e.g. medium access control [MAC] addresses · CPC title