Re: softirq considered harmful

[Jens Axboe <axboe@suse.de>]

On Sat, Aug 12 2006, Andrew Morton wrote:
> On Sat, 12 Aug 2006 18:09:44 -0700 (PDT)
> David Miller <davem@davemloft.net> wrote:
> 
> > From: Andrew Morton <akpm@osdl.org>
> > Date: Sat, 12 Aug 2006 17:45:49 -0700
> > 
> > > Is that also adding 150 usecs to each IO operation?
> > 
> > I have no idea, Jens hasn't done enough to narrow down the true cause
> > of the latencies he is seeing.  So pinpointing it on anything specific
> > is highly premature at this stage.
> 
> Determining whether pre-conversion scsi was impacted in the same manner
> would be part of that pinpointing process.
> 
> Deferring to softirq _has_ to add latency and any latency addition in
> synchronous disk IO is very bad.  That being said, 150 usecs per request is
> so bad that I'd be suspecting that it's not affecting most people, else
> we'd have heard.

Hopefully you often end up doing > 1 request for a busy IO sub system,
otherwise the softirq stuff is pointless. But it's still pretty bad for
single requests.

> > My point was merely to encourage you to find out the facts before
> > tossing accusations around. :-)
> 
> No, your point was that slotting this change into mainline without telling
> anyone was OK because SCSI has been doing something similar.

Not similar, identical. Andrew, there was _no_ real change there!

-- 
Jens Axboe