On Mon, Mar 31, 2014 at 09:59:22AM +0100, James Harrison wrote: [...] I wasn't aware that cache=none was necessary for live migration. It seems really broken that this is necessary, see Stefan's explanation here:
https://groups.google.com/forum/#!topic/synnefo-devel/9MU5ujOcOt4 > Also Caching=None is set because there is a disk cache on the disk. The qemu cache mode isn't really relevant to this. But I guess if your disk has a huge cache (let's say it's a SAN LUN or a very expensive RAID controller) then enabling the host cache (ie. cache=writeback) might be overkill. > Would double caching would slow down I/O? Not in my testing. cache=writeback is a little faster. There are some caveats: - cache=writeback is only safe if your guest sends write barriers, ie. recent Linux and recent Windows guests are OK, Linux and Windows prior to ~2008/2009 are NOT safe. - Our performance team reckon that cache=none is better than cache=writeback, but their test configuration fully provisions host memory with guests. You only benefit if your host actually has some free memory to give to the cache ... - It may make migration impossible. - Libguestfs is not representative of what real guests do. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top _______________________________________________ virt mailing list virt@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/virt
