Re: RBD snapshots cause disproportionate performance degradation

[Will Bryant <will.bryant@xxxxxxxxx>]

On 19/11/2015, at 23:36 , Haomai Wang <haomaiwang@xxxxxxxxx> wrote:

Hmm, what's the actual capacity usage in this volume? Fiemap could
help a lot to a normal workload volume like sparse data distribution.

I’m using basically the whole volume, so it’s not really sparse.

Hmm, it's really a strange result for me. AFAR, it should be a burst
bandwidth at least. From my tests, the disk bandwidth is keeping high
level.

What's your ceph env?

I have two; the production cluster is 9 OSD hosts with Crucial SSDs.  It is where we first observed the performance problem after taking snapshots.

To try and debug the problem I have built a small test cluster using an Intel S3710 SSD and 3 OSDs in containers on the same host.  This is not a very realistic deployment scenario, but it showed the same performance behaviour (and I think it eliminates the network latency as a source of problems).

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com