10GB Configuration.

[compuls1v3]

So I've configured 3 nodes, with 2 slow spinning disks, and 2 SSDs per host. There is one Ceph group, for the SSDs. Each server has 4x1GB NICs, with one port configured for management. They also have 1 dual 10GB NIC per server, configured for the cluster network. My question is, when I do a migration (with insecure mode enabled), should I be getting faster speeds than this?


Also, how do I change the migration cachesize to be bigger?

 

[t.lamprecht]

You seem to get around 320 MiB/s roughly averaged, multiplied with 8 (bits per byte) that would be ~ 2.5 Gbps, so if that is the upper limit of what's practical possible depends on what network you configured as migration-network (Datacenter -> Options) and what other traffic is happening there.

Note that available memory and CPU speed for QEMU's guest-state serialization resources may also have quite some impact and can be the actual limiting factor.

 

[compuls1v3]

Thank you for the reply. I changed the setting Maximal Works from 4 > 64, and now it transfers within a few seconds.

 

[t.lamprecht]

That makes honestly no sense to me. The worker settings is only used to determine how much parallel start/stop/migrate actions can be done by the HA stack or when triggering a bulk-action, but it does not affect a single migration itself at all.