Mark, If it's any help, we've done a small totally unreliable benchmark on our end. For a KVM instance, we had: 260MB/s write, 200MB/s read on local SAS disks, attached as LVM LVs, 250MB/s write, 90MB/s read on RBD, 32 osds, all SATA.
All sequential, a 10G network. It's more than enough currently but we'd like to improve RBD read performance. Cheers, On Sat, Mar 9, 2013 at 7:27 AM, Andrew Thrift <[email protected]>wrote: > Mark, > > > I would just like to add, we too are seeing the same behavior with > QEMU/KVM/RBD. Maybe it is a common symptom of high IO with this setup. > > > > Regards, > > > > > > Andrew > > > On 3/8/2013 12:46 AM, Mark Nelson wrote: > >> On 03/07/2013 05:10 AM, Wolfgang Hennerbichler wrote: >> >>> >>> >>> On 03/06/2013 02:31 PM, Mark Nelson wrote: >>> t >>> >>>> If you are doing sequential reads, you may benefit by increasing the >>>> read_ahead_kb value for each device in /sys/block/<device>/queue on the >>>> OSD hosts. >>>> >>> >>> Thanks, that didn't really help. It seems the VM has to handle too much >>> I/O, even the mouse-cursor is jerking over the screen when connecting >>> via vnc. I guess this is the wrong list, but it has somehow to do with >>> librbd in connection with kvm, as the same machine on LVM works just ok. >>> >> >> Thanks for the heads up Wolfgang. I'm going to be looking into QEMU/KVM >> RBD performance in the coming weeks so I'll try to watch out for this >> behaviour. >> >> >>> Wolfgang >>> ______________________________**_________________ >>> ceph-users mailing list >>> [email protected] >>> http://lists.ceph.com/**listinfo.cgi/ceph-users-ceph.**com<http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com> >>> >>> > > ______________________________**_________________ > ceph-users mailing list > [email protected] > http://lists.ceph.com/**listinfo.cgi/ceph-users-ceph.**com<http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com> > -- erdem agaoglu
_______________________________________________ ceph-users mailing list [email protected] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
