[Stgt-devel] Different throughput numbers on SLES 10 and RHEL 5.1

[Eli Dorfman]

On Sun, Mar 2, 2008 at 11:50 AM, Erez Zilber <erezz at voltaire.com> wrote:
> Mike Christie wrote:
>  > Erez Zilber wrote:
>  >> Hi,
>  >>
>  >> I'm using the latest stgt release (tgt-20071227) and I see different
>  >> throughput numbers for READ commands on SLES 10  & RHEL 5.1. I'm using
>  >> stgt with iSCSI over iSER.
>  >>
>  >> I'm using the same initiator (open-iscsi 865.15 from OFED 1.3 rc4) and
>  >> the same target machine. The target machine has 2 partitions - one is
>  >> SLES 10 & the other is RHEL 5.1. When I run READ commands against the
>  >> target on SLES 10, I get ~400 MB/sec (which is the throughput of the
>  >> LUN). When I run the same READ commands from the same initiator against
>  >> the target on RHEL 5.1, I get only 240 MB/sec.
>  >>
>  >> Is there anything that I need to config on the RHEL 5.1 machine? BTW -
>  >> if I run the same test directly from the target machine, I get 400
>  >> MB/sec on both distros.
>  >>
>  >> Here's the command that I use:
>  >>
>  >> sgp_dd if=/dev/sdc of=/dev/null bs=512 bpt=1024 thr=8 time=1
>  >> count=20480000
>  >>
>  >
>  > Does this command try to keep 8 IOs running in parrallel, and if so
>  > did you try the noop scheduler (I think for SG_IO this will not make a
>  > difference)?
>
>  I tried noop & cfq, but didn't see any dramatic change. BTW -
>  if=/dev/sg3 (not /dev/sdc), so the command that I'm running is:
>
>  sgp_dd if=/dev/sg3 of=/dev/null bs=512 bpt=1024 thr=8 time=1 count=20480000
>
>
>  >
>  > Is the target by any chance running on a x86_64 box?
>
>  Yes. Which box are you using? BTW - our non-stgt target is also running
>  on a x86_64 box successfully.
>
>  Erez
>
>
Mike,

Do you have any update (other ideas) on this issue?

Eli