MLAG (Multi-Chassis LAG), is a type of Link Aggregation Group with constituent ports that terminate on two separate switches and thus provide switch-level redundancy.
Why should I use LAG?
- To increase bandwidth from the server from single interface of 10/40GbE to two interfaces of such speeds
- To increase the bandwidth of a network link between switches without sacrificing the number of links due to spanning tree (STP) implications
- To get Top-of-rack (ToR) high availability (HA) for servers while allowing full use of the bandwidth of both links
- To achieve HA on a switch level without the consequences of STP
How is MLAG related to Mellanox VMS reference design?
- MLAG is a complimentary solution for the VMS reference design and can be configured in several locations within the design.
- MLAG can be configured at the ToR level and allow HA from the hosts while connecting uplinks from both switches towards several VMS leafs.
- MLAG can be configured on the VMS leafs and provide HA towards servers or switches without the bandwidth implication of using STP.
What are the limitation for the MLAG feature?
- In the maximum switch set, we have 64 ports, assuming at least one port is configured as IPL, the rest of the 63 ports can be configured as 63 MLAG interfaces
- Only one MLAG domain is supported per switch.
- The maximum number of switches that could be mapped to MLAG domain is two.
Is MLAG a L2 or a L3 solution?
MLAG is an L2 protocol and provides L2 connectivity with HA towards a peering device.
The uplink from the MLAG pair can be connected towards an L2 or L3 network. In the case of a L3 uplink, the MLAG implementation also provides routing functionality and default GW redundancy towards the L2 domain to which it is connected.
How do I configure MLAG?
Follow How To Configure MLAG on Mellanox Switches procedure.
MLAG is configured on the ToRs (or leafs)
There are two options to configure it depending on the size of the VMS.
- Small size VMS (consist of two spines)
- Large size VMS configured with L3
In a small size VMS, where there are two spines in the setup, MLAG can be configured in each two leafs as well as in the spines. In this case, all the network is configured as Ethernet L2, and there is no need for spanning-tree.
In large scale VMS topologies where L3 is enabled, MLAG can be configured on the VMS leafs to reach high availability on the ToRs. In this case, router redundancy protocol such as VRRP (or MAGP) is needed on the ToR uplink ports to conclude the configuration.