On Sat, 21 Feb 2009, Keith Freedman wrote: > is direct, a locally attached hard drive? > a network filesystem will NEVER perform as well as a locally attached disk. Odd, actually you should be able to make it preform BETTER, as you scale Lustre the I/O for even a single file scales. > I think you're numbers aren't too unreasonable. > You could probably improve your performance by adding some performance > translators. I have them on the server, should I also put them on the client? > Write behind would likely help you a bit. Also, even DRBD is better and it is IPoIP, not raw infiniband: [root at xen0 share]# dd if=/dev/zero of=/share/bar bs=1G count=8 8+0 records in 8+0 records out 8589934592 bytes (8.6 GB) copied, 60.8988 seconds, 141 MB/s > At 06:59 PM 2/21/2009, Nathan Stratton wrote: > >> Direct: >> [root at xen0 unify]# dd if=/dev/zero of=/sdb2/bar bs=1G count=8 >> 8+0 records in >> 8+0 records out >> 8589934592 bytes (8.6 GB) copied, 51.3145 seconds, 167 MB/s >> >> Gluster: >> [root at xen0 unify]# dd if=/dev/zero of=/unify/foo bs=1G count=8 >> 8+0 records in >> 8+0 records out >> 8589934592 bytes (8.6 GB) copied, 87.7885 seconds, 97.8 MB/s >> >> Boxes are connected with 10 gig Infiniband so that should not be an issue. >> >> http://share.robotics.net/glusterfs.vol >> http://share.robotics.net/glusterfsd.vol >> >>> <> >> Nathan Stratton CTO, BlinkMind, Inc. >> nathan at robotics.net nathan at blinkmind.com >> http://www.robotics.net http://www.blinkmind.com >> >> >> _______________________________________________ >> Gluster-users mailing list >> Gluster-users at gluster.org >> http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users >
