[ofa-general] RE: [Scst-devel] SRP aggregate bandwidth decreasing as threads increase

[Chris Worley]

On Mon, May 18, 2009 at 11:22 AM, Bart Van Assche
<bart.vanassche at gmail.com> wrote:
> On Mon, May 18, 2009 at 6:52 PM, Sufficool, Stanley
> <ssufficool at rov.sbcounty.gov> wrote:
>> IIRC, The SRP Target code has many context switches that throttle
>> performance at higher thread counts.
>
> Depends on which version of ib_srpt you are using. The ib_srpt kernel
> module has a parameter called "thread" which allows to control whether
> disk I/O is handled in another thread than the one that communicates
> over InfiniBand (thread=1) or in the same thread (thread=0). For older
> versions of the ib_srpt kernel module the default was thread=1, which
> caused indeed a lot of context switches. On December 3, 2008 (SCST
> Subversion revision 594) the default has been changed from thread=1 to
> thread=0 because the latter results in better performance.

I won't have access to the targets until tomorrow (at which point I
may not have internet access), so I'm trying to gather a few possible
solutions today.

I'm using a very recent version of the SCST target code, it would only
be ~1 month old.  So, I'm guessing I have the "thread=0" code.  Maybe,
for a high thread count, this needs to be "=1"?

Is there a way to control the number of threads once "thread=1" is
set?  Does it spawn one thread per initiator thread?

Any other ideas of things to try?

Thanks,

Chris
>
> Bart.
>