On 2009-10-30, at 05:49, udara weerapperuma wrote: > I have SLES 11 x86-64 and lustre v 1.8.1.1 installed in it. This > use as MDS and OSS. > > /dev/sda3 ----2 GB------ MDT > /dev/sdb2 ----5 GB------ OST1 > /dev/sdb3 ----5 GB------ OST2 > /dev/sdb5 ----5 GB------ OST3 > /dev/sdb6 ----5 GB------ OST4 > > Then I have 2 clients running on SLES 11 x86-64 and SLES 10 SP2 > x86-64 both has the lustre v 1.8.1.1 installed. > The SLES 10 client gives the same speed for every bs value in dd > command (dd if=/dev/zero of=/lustre/sles10.out bs=1k gives 43 MB/s) > This is same for the every bs value on SLES 10 SP2. > > But for the SLES 11 x86-64 client gives only 4 MB/s speed for the > small bs values. (bs=1k) It gives 45 MB/s for the bs=32k values and > up.
Using multiple partitions on the same disk is a sure way to get very bad performance. Secondly, bs=1k is also a good way to add a lot of overhead to the IO. At a minimum use bs=4k (to match the client page size). Cheers, Andreas -- Andreas Dilger Sr. Staff Engineer, Lustre Group Sun Microsystems of Canada, Inc. _______________________________________________ Lustre-discuss mailing list [email protected] http://lists.lustre.org/mailman/listinfo/lustre-discuss
