On 10.08.2012 13:46, Julian price wrote: > I have 2 similar servers. Since upgrading one from CentOS 5.5 to 6, > disk write performance in kvm guest VMs is much worse. > > There are many, many posts about optimising kvm, many mentioning disk > performance in CentOS 5 vs 6. I've tried various changes to speed up > write performance, but northing's made a significant difference so > far: > > - Install virtio disk drivers in guest > - update the host software > - Update RAID firmware to latest version > - Switch the host disk scheduler to deadline > - Increase host RAM from 8GB to 24GB > - Increase guest RAM from 2GB to 4GB > - Try different kvm cache options > - Switch host from ext4 back to ext3 > - Set noatime on the virtual disk image file > Note: There is no encryption or on-access virus scanner on any host > or guest. > > Below are some the block write figures in MB/s from bonnie++ with > various configurations: > > First, figures for the hosts show that the CentOS 6 server is faster: > > 54 CentOS 5 Host > 50 CentOS 5 Host > 69 CentOS 6 host > 70 CentOS 6 host > > Figures for a CentOS 6 guest running on the CentOS 5 host show that > the performance hit is less than 50%: > > 30 CentOS 6 guest on CentOS 5 host with no optimisations > 27 CentOS 6 guest on CentOS 5 host with no optimisations > 32 CentOS 6 guest on CentOS 5 host with no optimisations > > Here are the figures a CentOS 6 guest running on the CentOS 6 host > with various optimisations. Even with these optimisations, > performance doesn't come close to the un-optimised guest running on > the CentoOS 5 host: > > 5 No optimisations (i.e. same configuration as on CentOS 5) > 4 deadline scheduler > 5 deadline scheduler > 15 noatime,nodiratime > 14 noatime,nodiratime > 15 noatime > 15 noatime + deadline scheduler > 13 virtio > 13 virtio > 10 virtio + noatime > 9 virtio + noatime > > The CentOS 6 server has a better RAID card, different disks and more > RAM, which might account for the better CentOS 6 host performance. > But why might the guest write performance be so much worse? > > Is this a known problem? If so, what's the cause? If not, is > there a way to locate the problem rather than using trial and error? > > Thanks, > Julian How are you running KVM machines? Through libvirt? Have you made any changes to the cgroups? What are you using as disks, raw files, lvm? -- Sent from the Delta quadrant using Borg technology! Nux! www.nux.ro _______________________________________________ CentOS-virt mailing list CentOS-virt@xxxxxxxxxx http://lists.centos.org/mailman/listinfo/centos-virt
