Re: [Users] Fedora 20

> Von: Itamar Heim [iheim(a)redhat.com] > Gesendet: Sonntag, 24. November 2013 11:54 > An: Markus Stockhausen; users(a)ovirt.org > Betreff: Re: AW: [Users] Fedora 20 >> >> Being on a current kernel in the hypervisor just gives a better >> feeling in a 10GBit NFS environment. That comes from several >> "would-have-been-nice-to-have-that-optimization-option" >> experiences. >> > > interesting. i assume you mean for the client machines, rather than the > NFS server. > to learn a bit more on this - do you have some examples of such > optimizations which are in latest fedora but not in say, el6.5 kernel? > The feeling is not related to a current comparison but to past situations where a new linux kernel provided some nice feature to squeeze more performance out of the existing infrastructure. We have a lot of SAP test/demo systems that are usually >100GB. Moving data around or (un)loading databases can be quite a headache. Especially with respect to other running systems. Some examples that were helpful in our environment: - bufferbloat fighting patches (3.6 and on) - Using SMBv2 client in the kernel (since 3.7). E.g. direct access to a windows share on the hypervisor - perf patches for better bottleneck analysis Going with Ovirt and the possibility to do more things online (relocating VMs and their disks) may reduce the requirement for the newest kernel version. So starting with Fedora 19 on the nodes and a more stable CentOS on the engine could be a good tradeoff. What do you think?