On 02/15/2011 11:15 AM, James Masson wrote:
I wouldn't use Raid5/6 take a look at these NFS stats - my VM hosting workload is 90% write. For my everyday VM workloads, the only time there are significant reads from the VM shared storage is at VM boot time, and even then, the VM host and storage server have a significant parts of the VM disks in cache. Server rpc stats: calls badcalls badauth badclnt xdrcall 923254581 0 0 0 0 Server nfs v3: null getattr setattr lookup access readlink 32 0% 761649 0% 161 0% 461 0% 112220 0% 0 0% read write create mkdir symlink mknod 24842856 2% 838746432 90% 183 0% 9 0% 2 0% 0 0% remove rmdir rename link readdir readdirplus 120 0% 1 0% 73 0% 0 0% 0 0% 289 0% fsstat fsinfo pathconf commit 10248397 1% 32 0% 0 0% 48541412 5%
Yes I guess a lot of the reading I/O will be caught be the caches so write performance becomes the main issue. I mostly threw the raid-5 in there to have some basic redundancy on the individual nodes so you don't have to do a full sync of the drbd device when a single disk dies. Raid-10 seemed wasteful to me but I guess given todays disk prices and the fact that even with 1TB HD's you'd still get 4TB of usable storage that looks like the better option. In any case I'm going to do some benchmarking on the setup once I get my hands on it to get some hard numbers.
Regards, Dennis _______________________________________________ drbd-user mailing list drbd-user@lists.linbit.com http://lists.linbit.com/mailman/listinfo/drbd-user
