-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Le 19/02/2015 19:20, Alexandre DERUMIER a écrit : >>> I don't think it's a good idea to build Ceph on top of ZFS. >>> Ceph expects to give it full hard disks for storage and using >>> ZFS with multiple disks will degrade performance. > > Yes, don't use zfs for ceph. It's not officially supported, and as > ceph use specific features of each filesystem (xfs, ext4 and > btrfs), I can't recommand to use zfs in production. > > xfs is the current recommended filesystem for ceph osds. > > OK, ok, i surrender. Read too fast the docs i found. XFS instead seems to be better.
But, even if i understand the mandatory drawbacks of a clustered FS (ceph), i find the performance penalty to be huge : According to http://www.sebastien-han.fr/blog/2012/08/26/ceph-benchmarks/ with the "Insane servers" (close to mine) and a dedicated 1G network (mine will be 10gbe), the write bandwidth is between 110 and 150 MB/s. It's not a shame, but with 64GB RAM, Raid of SSD and 32 Cores, you could expect twice the bandwidth at least. OK, for the redundant FS, but what a huge price i'd pay for.. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iEYEARECAAYFAlTmMDYACgkQlhqCFkbqHRa+1QCgrOwHMNWzQhM+h+7jYR+52ORb JLsAoM3GoI/3y1YdpGczrSGpanqtpG95 =yBZv -----END PGP SIGNATURE----- _______________________________________________ pve-user mailing list [email protected] http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
