[PVE-User] High rate of Multicast traffic between cluster nodes

[Andrew Thrift]

Hi,

We have a PVE 4.4 cluster with 16x nodes and around 600 VM's.   We
have noticed a lot of errors in the logs relating to corosync
re-transmits, e.g.:

vmh10 corosync[29335]: [TOTEM ] Retransmit List: 514b4d 514b51 514b58
514b5a 514b5b 514b5d 514b5f

Upon investigation we found that the number of multicast packets was
exceeding the limit of multicast storm protection on our switch ports
(set to 1000pps), we raised this up to 16000pps per port and were
still seeing corosync re-transmits.  Disabling all multicast storm
protection eliminates the problem completely.

Why is the rate of multicast so high from Corosync ? e.g. is it by design ?
Is there anyway to reduce this ?


Thanks,



Andrew