Hi list. Im Running xen on a brand new dell PE 1950 with 146 GB SAS Disk on raid1. The performance on the I/O on the domU is really poor, but in dom0 the performance is great. Any ideas? Regards Francisco. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.centos.org/pipermail/centos-virt/attachments/20090519/7a9dc81d/attachment-0006.html>
Francisco P?rez <fperez.x at gmail.com> writes:> Hi list. > > Im Running xen on a brand new dell PE 1950 with 146 GB SAS Disk on raid1. > The performance on the I/O on the domU is really poor, but in dom0 the > performance is great.Are you using HVM?
----- "Francisco P?rez" <fperez.x at gmail.com> wrote:> OK, here it is: > > name = "vm01" > uuid = "2fa05fe5-9b52-3527-5bf5-efb88555c567" > maxmem = 5836 > memory = 5836 > vcpus = 1 > bootloader = "/usr/bin/pygrub" > on_poweroff = "destroy" > on_reboot = "restart" > on_crash = "restart" > vfb = [ ] > disk = [ "tap:aio:/var/lib/xen/images/vm01.img,xvda,w" ] > vif = [ "mac=00:16:3e:15:6f:cd,bridge=xenbr0" ] > > Thanks.What kind of "disk performance"? Read? Write? Random? Linear? As for general tuning, use block devices (LVM) unless you have to do otherwise. Do you do a large amount of I/O in general? Do you have dom0's CPU pinned to an unshared core? -- Christopher G. Stach II
----- "Francisco P?rez" <fperez.x at gmail.com> wrote:> I guess the bad performance it goes with the writing. For example on > decompression of a big tar.gz, installing rpm packages. Also on > restoring a MySQL Database... > > The first clue when i knew that something were wrong was on the FS > formating stage. It took really long time to format 40 GB to ext3. I > did the same on other machine with a sata disk... surprise! same > software and hardware config except for the inferior disk type and did > it in no time.It sounds like your the virtual disk file is sparse. If that's the case, you traded performance for space. It not only has to deal with the underlying filesystem, but it allocates the space on demand and hits that filesystem's journal. Try an LVM logical disk and see how that performs. -- Christopher G. Stach II
----- "Francisco P?rez" <fperez.x at gmail.com> wrote:> Thanks all of you. Just with the nonsparse option works all right. > > Then an LVM+nonsparse+phy will be the better setup? > > Regards.Yes, except sparseness is only relevant to file-backed vbds in normal use. (There could ostensibly be a similar lazy block allocation option for a target on something like a SAN/NAS and you would expect some performance degradation for some writes.) -- Christopher G. Stach II