Live Migration Very Slow
-
Xen Orchestra, commit 32228
Master, commit 32228Hi,
I have a two host pool that uses shared storage for all VMs. When I try to perform a VM migrate, using either the management network (1Gbs) or the 20Gbps NFS network, the live migration can take a long time (see example below):
I thought when the VM is using the shared storage of a pool, then moving VMs should be very fast as there is no disk data that needs to be moved other than the VM ram.
Both hosts Control Domains have assigned 16 GB of memory.
Source host shows the following:
Target host shows the following:
Any recommendations on what could be causing the slow live migration if the VM resides on the pool shared storage?
Also, our XOCE VM resides on a different pool than the pool I'm performing the live migration. Wasn't sure if that matters or not.
Thank You,
SW
-
Were you able to work it out? I have a 32GB RAM domU that keeps failing migration between nodes unfortunately!
-
@andrewperry I couldn't get the live migration to work on large VMs. To move large VMs, I ended up taking a snapshot of the VM and then creating a new VM from that snapshot. It's not ideal but had to do that as we needed to retire some old servers and I couldn't get the live migration to work w/ large VMs.
-
@stevewest15 that is a shame! Thanks for sharing!
-
Live migration speed per VM isn't really fast due how memory is split is very small 4k blocks. However, it scales OK if you migrate multiple VMs at once. Obviously, this is not a solution in your case where you need faster migration per VM. This is a topic we'd like to work on in the following months
-
Hi, sorry for revisiting an older topic, but we have the same issue with slow VM migration. We changed from 2x1G to 2x10G network, but the migration performance from one host's local SR to another host local SR is not much improved.
Using XCP-ng 8.2.1 with up-to-date patches. Local storage on both hosts is SSD RAID1, ext4.
It would be very good if we could improve this situation. Currently we are seeing only 5% network utilisation.
-
@Forza Sorry but that's unrelated in this topic. OP talks about live migration with a shared SR. Storage migration is really another topic.
-
@olivierlambert aha, I misinderstood. Should I open another topic or perhaps a support ticket?
-
@Forza There are various topics touching on how SMAPIv1 is a bottleneck here, eg: https://xcp-ng.org/forum/topic/9389/backup-migration-performance/8 - that's probably not the best example as it is also more on backups than migration!
