The LACP rate fast feature is used to set the rate (once every second) at which the LACP control packets are sent from partner. The normal rate at which LACP packets are sent is 30 seconds.
This document provides workflow of the LACP rate fast feature including the packet capture and some recommendations/concerns in MLAG setup.
How it works
When LACP is synchronizing between two device LACPDUs are sent at a rate of 1 per second until both sides are synchronized. Once this is complete they are sent at a rate of 1 per 30 seconds. LACP Rate Fast is a command that allows partners to send LACPDUs at a rate of 1 per second even after synchronization.
By setting a bit or flag in the LACPDU, which is known as the LACP Timeout bit. When an LACP aware device receives this flag/bit it will then begin sending at a rate of once per second rather than once per 30 seconds.
It is important to note that setting this command locally on an Arista device will not affect the rate the Arista is sending at. It will only affect the rate of the remote device. Therefore configuring ‘lacp rate fast’ locally really means ‘lacp receive fast’.
CLI show commands
This output is used to see the Actor state.
------------- show lacp internal detailed all-ports ------------- LACP System-identifier: 8000,74-83-ef-d7-60-a9 MLAG System-identifier: 8000,76-83-ef-d7-60-a9 State: A = Active, P = Passive; S=ShortTimeout, L=LongTimeout;
This output is used to see the Partner state.
------------- show lacp neighbor detailed all-ports ------------- State: A = Active, P = Passive; S=ShortTimeout, L=LongTimeout;
Wireshark output of LACPDU Flags
The below LACPDU is sent from Nortel to the Arista. Nortel is the actor and the Arista is the partner. The Actor State has LACP Timeout as “No” which means the Nortel is not requesting LACPDU at a fast rate, in other words it does not have a Short Timeout for LACPDU messages and therefore the Arista will send once per 30 seconds. The Partner State – this refers to the Arista – shows that LACP Timeout flag is set to “Yes”. This means that the Arista is requesting the Nortel to send at a fast rate. In other words the Arista device has a fast LACP Timeout and requires the Nortel to send at a rate of once per second. This state will occur if the Arista has ‘lacp rate fast’ enabled on the interface while the remote device does not have this configured.
We strongly recommend to *not* run “lacp rate fast” over the peerlink port-channel. LACP, being a control plane protocol, can be affected by other features that could take up CPU cycles.
For small scale MLAG + lacp rate fast deployments on MLAG port-channels should work fine. That said, we would highly recommend testing “lacp rate fast” in a non-production setup first. Once configured, you can run “top” to see CPU usage of the system and specifically, LacpTxAgent. The CPU usage for LacpTxAgent should be in the low single digits. Once we are confident that the configuration is stable, then we can apply to production.
For large scale MLAG domains, the LACP rate fast should not be used. For example, Modular devices with many ports. Please consult your SE for the design choices or concerns.