Quoting Serge van Ginderachter ([email protected]): > On 2 April 2011 16:58, Clint Byrum <[email protected]> wrote:
> Some general conclusions were: > > * VMWare ESXi was the clear winner, > * Xen on Debian Lenny was a close second, and the winner amongst Open > Source solutions. > * Xen performance is way better than KVM, especially when looking at disk > access > * Xen performs in a more stable and predictable way, whilst KVM seemed > to perform more at random (which confirms Iustin's observations, ) > * CentOS (5.4) performed remarkably well for being older sofwtare > versions (KVM, Xen, Linux kernel) > * performance on Ubuntu was really bad. The then recent Ubuntu Lucid > was far worse than CentOS 5.5 (both KVM) On the one hand, you can't make claims like this without giving very detailed info on the storage configuration. On the other hand, kvm (and libvirt) tend to make changes which impact how you need to tune things to get best performance. Which is not really acceptable in a real enterprise deployment. We've gotten complaints before - valid IMHO - about 'undocumented changes' like that. This is IMO a strong consideration for considering re-enabling xen. It also may deserve a UDS topic on whethere there is something we can do. Perhaps we can spend a week around alpha-3 time doing performance tests of various configurations. Perhaps we can query the community for what they consider current best practices, and document those at release time. Perhaps query, then do our week of performance tests to validate, then document. -serge
signature.asc
Description: Digital signature
-- ubuntu-server mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-server More info: https://wiki.ubuntu.com/ServerTeam
