[Stgt-devel] sg_turs on stgt iscsi drive is very slow

[Ming Zhang]

On Sun, 2007-01-21 at 00:47 +0900, FUJITA Tomonori wrote:
> From: Ming Zhang <blackmagic02881 at gmail.com>
> Subject: Re: [Stgt-devel] sg_turs on stgt iscsi drive is very slow
> Date: Sat, 20 Jan 2007 10:30:58 -0500
> 
> > On Sun, 2007-01-21 at 00:22 +0900, FUJITA Tomonori wrote:
> > > From: Ming Zhang <blackmagic02881 at gmail.com>
> > > Subject: [Stgt-devel] sg_turs on stgt iscsi drive is very slow
> > > Date: Mon, 11 Dec 2006 11:58:27 -0500
> > > 
> > > > Pulled out the linux-2.6-target git tree, compiled stgt and iet under
> > > > same kernel. Export a 4 disk raid0 device to ini. Ini is open-iscsi
> > > > r598.
> > > > 
> > > > Want to know how efficient stgt handle context switch. 
> > > > 
> > > > run  sg_turs -t -n=10000 /dev/sg0 on IET drive. results is 2521/sec.
> > > 
> > > (snip)
> > > 
> > > > run  sg_turs -t -n=10000 /dev/sg0 on stgt drive. results is 250
> > > > operations/sec. yes, 250.
> > > 
> > > OK. I did some tests too. As we know, aioepoll isn't effective.
> > > 
> > > The sg_turs test doesn't need AIO code. So I just tried epoll for
> > > event notification.
> > > 
> > > - tgt (patched)
> > > 
> > > paris:~# sg_turs -t -n=1000 /dev/sg3
> > > time to perform commands was 0.125131 secs; 7991.62 operations/sec
> > > Completed 1000 Test Unit Ready commands with 0 errors
> > > 
> > > - IET
> > > 
> > > paris:~# sg_turs -t -n=1000 /dev/sg3
> > > time to perform commands was 0.128747 secs; 7767.17 operations/sec
> > > Completed 1000 Test Unit Ready commands with 0 errors
> > > 
> > > 
> > > Well, not bad at all. And I heard that kevent is more effective than
> > > epoll.
> > 
> > the # looks pretty good. thanks for testing. maybe it is bounded by your
> > network now? could you try a open-iscsi via loopback?
> 
> For what do we need to know the performance in unreal environments? I
> access Linux boxes remotely. So I don't want to crash them.
> 
> Anyway, the patch is available so you can try.

sure.

> 
> 
> > > Note that aioepoll affects only user-mode target drivers (now only
> > > iSCSI). The kernel target drivers don't need AIO code (can use
> > > epoll). So they should have no performance problem. And as I said
> > > before, I think that the iSCSI target driver is ok in real workloads.
> > 
> > do you have a road map for stgt? see, need to read your reply for Pete
> > to know OSD is already under your plan.
> 
> Well, I decided not to plan anything. I was supposed to finish iSER
> target code however I'm playing with other stuff like sg v4.

this does not do good for stgt project in long term. u can always lay it
out and people know how to contribute more efficiently.


-- 
http://blackmagic02881.wordpress.com/