This is an interesting question I have recently faced. I guess the problem is solved at the hypervisor level, so the question is do all the I/O ops of the memory-bound VM have a greater weight than the stress on the CPU, or does the hypervisor simply stops the VM and any locks are instantaneously released. I would like to know whether or not the two VMs finish migration at the same time.
Hello Marc,
I wish to share my views on your question.
I think it varies with the type of Virtualization ,the Hypervisor and File System [variation between Source & Destination ex. NFS, LVM etc] .
Performance of Live Migration will be affected if NFS is used , due to the bandwidth limitation, I/O Latency between Source & Destination System.
virtualization is catetgorized as Full Virtualization & Para Virtualization.
For Live migration, Community suggest to have a shared storage which contains same copy of VM in the Nodes.I think Copying Memory Mapping & syncing across Source & Destination is a challenging task while live migration.
See the link, which discusses about live migration in Xen
https://www.suse.com/communities/conversations/xen-virtual-machine-migration/
http://wiki.prgmr.com/mediawiki/index.php/Chapter_9:_Xen_Migration#Live_Migration
If you consider only the volatile states (i.e.the RAM and not the VM image), you can give a look to the paper we published at CloudCom 2013. This paper discusses a live migration model based on several experiments showing that the live vm migation time is mainly governed by the memory update rate (i.e. how many memory pages are updated during the live migration).
http://dl.acm.org/citation.cfm?id=2568524
- Badges
- Science topic