On Fri, Apr 26, 2013 at 10:30:01AM +0000, James Harper wrote:
Converting back to xfs is something I'd been meaning to do for a while - the VM is only about 10GB is size so it only takes a few minutes to cp -a.
what kind of "disk" is the VM's disk running on? mdadm+lvm? what kind of raid?
Yes, LVM on RAID1 on 2 x 7200RPM SATA disks.
i'm wondering if you've lucked onto some pathological worst-case of striping and block size.
Could be. I'm puzzled as to how this problem spontaneously developed though, and if the problem is: . mysql suddenly deciding to fsync after each write to the innodb log file (which is what I've tracked it down to) . fsync suddenly going slow under my xen DomU . Dom0 suddenly not handling vbd ring flush operations well . something else... Not sure if I'm testing the right thing here, but the following gives me ~30kbytes/second on most machines I test this on: dd if=/dev/zero of=test.bin bs=512 count=128 oflag=sync (software raid obviously, hardware raid with battery backed cache should be about 10x-100x faster)
also do you have any/many snapshots of that LV? snapshots can really bugger up LVM performance.
No snapshots. Thanks James