Can a volume move effect the whole cluster?

[mattmusgrove1]

12-node 9.1P8 cluster

I ran a volume move which completed successfully in 4.5 hours, 137.7MB/s.  During this time and several hours afterwards I have had users report problems with NFS mounts however the issues the users were reporting were not using the nodes or aggregates where the volume move was taking place.  My question is, can a volume move effect performance elsewhere in the cluster?

Many thanks

[aladd · Accepted solution]

Hello @mattmusgrove1,

Depending on overall layout and workload, a volume move can absolutely impact performance.

Before performing a volume move, it may be suggested to know what your average usage is on the cluster and where any hot spots may be.

If I recall, I believe the volume move function uses the same engine as SnapMirror, and will use the cluster interconnect for transfer between nodes.

Do you know if your NFS traffic is direct (volume and LIF are hosted on the same node) or indirect? (volume an LIF live on different nodes).

[mattmusgrove1]

Thanks @aladd for your reply.  In the first example I looked at the LIF and volume being mounted were indeed on different nodes, different HA pairs in fact.  So in summary, if running a volume move utilizes the cluster interconnect then all other traffic in the cluster also using the interconnect could potentially be effected.  So from this do I conclude it is preferable in Linux when NFS mounting to use a LIF on the same node that the volume is located to avoid traffic using the interconnect?

[aladd]

Hello @mattmusgrove1,

This can be based purely on circumstances and your needs. We have a requirement for a 10G connection for cluster interconnect to facilitate this traffic, however depending on your environment and use of the system can have an impact.

[mattmusgrove1]

OK thanks @aladd .  I had a cluster port fail on the node hosting the NFS volume due to a high number of CRC errors so I suspect this was a contributing factor.