[Stgt-devel] Tuning iSER for performance

[Erez Zilber]

Robin Humble wrote:
> On Sun, Feb 10, 2008 at 05:19:03PM +0200, Erez Zilber wrote:
>   
>> Robin - did you try to use a real device? If yes, did you see any
>> difference between running something like sgp_dd directly from the
>> target machine and running the same sgp_dd command from the initiator
>> machine?
>>     
>
> I've used real devices for some tests in the past, but haven't ever
> tried initiator on the same machine. sorry.
> the tests that I can do easily are with backing store files on lustre,
> which also uses IB, so that halves the available IB bandwidth for iSER.
>
> I can try initiator on the same machine with lustre in a few days time.
> direct attached storage will prob have to wait a month 'til I get back
> from Canada :-/
>
> cheers,
> robin
>   

Robin,

I'm adding stgt list to this thread.

I'm not talking about running the initiator & the target from the same
machine. What I said is that if you're running the initiator on machine
'A' and the target on machine 'B', I see the following behavior:

    * running sgp_dd directly from the target machine - 400 MB/sec
      (which is the throughput of the LUN)
    * running sgp_dd from the initiator on the same LUN - 240 MB/sec

If you get better numbers, I'm interested to know if you're using the
default config or anything else.

Erez