On Feb. 06, 2008, 16:19 +0200, Miguel Gonzalez CastaƱos <[EMAIL PROTECTED]> wrote: > Hi, > > I'm trying to test iSCSI under VMWare.
Well, vmware is certainly one too many variable to test at the stage you are. You should try to establish a baseline measurement over a clean environment, i.e. a machine with controlled configuration running only the benchmark and the code under test. Then, if you want to measure vmware's effect on performance, repeat the test under vmware. I'd also recommend planning the experiments such the results will be comparable (e.g. simulating the same memory size on the reference system and vmware, similar network driver, etc.) > > I've been told to first test raw disk using disktest: > > reads > > disktest -PT -T30 -K32 -B128k -ID /dev/sdXYZ -h1 > > writes > > disktest -PT -T30 -K32 -B128k -ID /dev/sdXYZ -h1 -D 0:100 > > If I use this configuration I get around 90 MB/s for writes and 1900 > MB/s for reads. If I add the flag -If then is meant to fsync after each > write (this deactivates the cache, right?) then the performances lows to The cache is not deactivated, but rather data is flushed to storage on fsync(), so you're effectively not caching any data for write-behind. > 5 MB/s for writes and 34 MB/s for reads I'm not sure why read throughput dropped that much though as fsync() is not supposed to invalidate the cache. > > To compare, the local drive (not the iSCSI drive), I get using cache > 374.60MB/s in writes and 2290.25MB/s in reads and without cache 8.10MB/s > for writes and 30.12MB/s for reads. Cached results (on the initiator) are irrelevant if you want to measure iSCSI's performance. > > When I use dd for filesystem benchmarking What are you mounting exactly? > > for writes: > > dd if=/dev/zero of=/mnt/raid1/zero bs=4096 count=1572864 > > or > > dd if=/dev/zero of=/mnt/raid1/zero bs=128k count=52428 > > and reads: > > dd if=/mnt/virtualdata1/zero of=/dev/null bs=4096 > > or > > dd if=/mnt/raid1/zero of=/dev/null bs=128k > > I enabled noop scheduler: > > echo noop > /sys/block/sdc/queue/scheduler > > The results that I get from dd are much closer to the disktest using > fsync. I've enabled noatime in the mounting options but I don't get > better results. > > am I getting reasonable results? That's hard to say under vmware. Benny > > Miguel > > > > > > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~----------~----~----~----~------~----~------~--~---