From: Jens Axboe <axboe@suse.de>
To: Chris Mason <mason@suse.com>
Cc: Marcelo Tosatti <marcelo@conectiva.com.br>,
lkml <linux-kernel@vger.kernel.org>,
"Stephen C. Tweedie" <sct@redhat.com>,
Alan Cox <alan@lxorguk.ukuu.org.uk>,
Jeff Garzik <jgarzik@pobox.com>, Andrew Morton <akpm@digeo.com>,
Andrea Arcangeli <andrea@suse.de>,
Alexander Viro <viro@math.psu.edu>
Subject: Re: RFC on io-stalls patch
Date: Sun, 13 Jul 2003 19:47:09 +0200 [thread overview]
Message-ID: <20030713174709.GA843@suse.de> (raw)
In-Reply-To: <1058113238.13313.127.camel@tiny.suse.com>
On Sun, Jul 13 2003, Chris Mason wrote:
> On Sun, 2003-07-13 at 05:01, Jens Axboe wrote:
> > On Sat, Jul 12 2003, Chris Mason wrote:
> > > On Sat, 2003-07-12 at 03:37, Jens Axboe wrote:
> > >
> > > > > I believe the new way provides better overall read performance in the
> > > > > presence of lots of writes.
> > > >
> > > > I fail to see the logic in that. Reads are now treated fairly wrt
> > > > writes, but it would be really easy to let writes consume the entire
> > > > capacity of the queue (be it all the requests, or just going oversized).
> > > >
> > > > I think the oversized logic is flawed right now, and should only apply
> > > > to writes. Always let reads get through. And don't let writes consume
> > > > the last 1/8th of the requests, or something like that at least. I'll
> > > > try and do a patch for pre4.
> > >
> > > If we don't apply oversized checks to reads, what keeps a big streaming
> > > reader from starving out all the writes?
> >
> > It's just so much easier to full the queue with writes than with reads.
> >
>
> Well, I'd say it's a more common problem to have lots of writes, but it
> is pretty easy to fill the queue with reads.
s/easy/occurs often. Bad wording, yes of course it's trivial to make it
happen. But on a desktop machine, it rarely does. Can't say the same
thing about writes.
> > > The current patch provides a relatively fixed amount of work to get a
> > > request, and I don't think we should allow that to be bypassed. We
> > > might want to add a special case for synchronous reads (like bread), via
> > > a b_state bit that tells the block layer an immediate unplug is coming
> > > soon. That way the block layer can ignore the oversized checks, grant a
> > > request and unplug right away, hopefully lowering the total number of
> > > unplugs the synchronous reader has to wait through.
> > >
> > > Anyway, if you've got doubts about the current patch, I'd be happy to
> > > run a specific benchmark you think will show the bad read
> > > characteristics.
> >
> > No I don't have anything specific, it just seems like a bad heuristic to
> > get rid of. I can try and do some testing tomorrow. I do feel strongly
> > that we should at least make sure to reserve a few requests for reads
> > exclusively, even if you don't agree with the oversized check. Anything
> > else really contradicts all the io testing we have done the past years
> > that shows how important it is to get a read in ASAP. And doing that in
> > the middle of 2.4.22-pre is a mistake imo, if you don't have numbers to
> > show that it doesn't matter for the quick service of reads.
>
> I believe elevator-lowlatency tries to solve the 'get a read in ASAP'
> from a different direction, by trying to limit both the time required to
> get a request and the time for required for the unplug to run. Most of
> my numbers so far have been timing reads in the face of lots of writes,
> where elevator-lowlatency is a big win.
>
> It should make sense to have a reserve of requests for reads, but I
> think this should be limited to the synchronous reads. Still, I haven't
> been playing with all of this for very long, so your ideas are much
> appreciated.
Just catering to the sync reads is probably good enough, and I think
that passing in that extra bit of information is done the best through
just a b_state bit.
I'll try and bench a bit tomorrow with that patched in.
--
Jens Axboe
next prev parent reply other threads:[~2003-07-13 17:46 UTC|newest]
Thread overview: 68+ messages / expand[flat|nested] mbox.gz Atom feed top
2003-07-08 20:06 RFC on io-stalls patch Marcelo Tosatti
2003-07-10 13:57 ` Jens Axboe
2003-07-11 14:13 ` Chris Mason
2003-07-12 0:20 ` Nick Piggin
2003-07-12 18:37 ` Chris Mason
2003-07-12 7:37 ` Jens Axboe
2003-07-12 7:48 ` Jens Axboe
2003-07-12 18:32 ` Chris Mason
2003-07-13 0:33 ` Andrea Arcangeli
2003-07-13 9:01 ` Jens Axboe
2003-07-13 16:20 ` Chris Mason
2003-07-13 16:45 ` Jeff Garzik
2003-07-13 19:33 ` Andrea Arcangeli
2003-07-13 17:47 ` Jens Axboe [this message]
2003-07-13 19:35 ` Andrea Arcangeli
2003-07-14 0:36 ` Chris Mason
2003-07-13 19:19 ` Andrea Arcangeli
2003-07-14 5:49 ` Jens Axboe
2003-07-14 12:23 ` Marcelo Tosatti
2003-07-14 13:12 ` Jens Axboe
2003-07-14 19:51 ` Jens Axboe
2003-07-14 20:09 ` Chris Mason
2003-07-14 20:19 ` Andrea Arcangeli
2003-07-14 21:24 ` Chris Mason
2003-07-15 5:46 ` Jens Axboe
2003-07-14 20:09 ` Marcelo Tosatti
2003-07-14 20:24 ` Andrea Arcangeli
2003-07-14 20:34 ` Chris Mason
2003-07-15 5:35 ` Jens Axboe
[not found] ` <20030714224528.GU16313@dualathlon.random>
2003-07-15 5:40 ` Jens Axboe
[not found] ` <1058229360.13317.364.camel@tiny.suse.com>
2003-07-15 5:43 ` Jens Axboe
[not found] ` <20030714175238.3eaddd9a.akpm@osdl.org>
[not found] ` <20030715020706.GC16313@dualathlon.random>
2003-07-15 5:45 ` Jens Axboe
2003-07-15 6:01 ` Andrea Arcangeli
2003-07-15 6:08 ` Jens Axboe
2003-07-15 7:03 ` Andrea Arcangeli
2003-07-15 8:28 ` Jens Axboe
2003-07-15 9:12 ` Chris Mason
2003-07-15 9:17 ` Jens Axboe
2003-07-15 9:18 ` Jens Axboe
2003-07-15 9:30 ` Chris Mason
2003-07-15 10:03 ` Andrea Arcangeli
2003-07-15 10:11 ` Jens Axboe
2003-07-15 14:18 ` Chris Mason
2003-07-15 14:29 ` Jens Axboe
2003-07-16 17:06 ` Chris Mason
2003-07-15 9:22 ` Chris Mason
2003-07-15 9:59 ` Andrea Arcangeli
2003-07-15 9:48 ` Andrea Arcangeli
2003-07-14 20:16 ` Andrea Arcangeli
2003-07-14 20:17 ` Marcelo Tosatti
2003-07-14 20:27 ` Andrea Arcangeli
2003-07-15 5:26 ` Jens Axboe
2003-07-15 5:48 ` Andrea Arcangeli
2003-07-15 6:01 ` Jens Axboe
2003-07-15 6:33 ` Andrea Arcangeli
2003-07-15 11:22 ` Alan Cox
2003-07-15 11:27 ` Jens Axboe
2003-07-16 12:43 ` Andrea Arcangeli
2003-07-16 12:46 ` Jens Axboe
2003-07-16 12:59 ` Andrea Arcangeli
2003-07-16 13:04 ` Jens Axboe
2003-07-16 13:11 ` Andrea Arcangeli
2003-07-16 13:21 ` Jens Axboe
2003-07-16 13:44 ` Andrea Arcangeli
2003-07-16 14:00 ` Jens Axboe
2003-07-16 14:24 ` Andrea Arcangeli
2003-07-16 16:49 ` Andrew Morton
2003-07-15 18:47 Shane Shrybman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20030713174709.GA843@suse.de \
--to=axboe@suse.de \
--cc=akpm@digeo.com \
--cc=alan@lxorguk.ukuu.org.uk \
--cc=andrea@suse.de \
--cc=jgarzik@pobox.com \
--cc=linux-kernel@vger.kernel.org \
--cc=marcelo@conectiva.com.br \
--cc=mason@suse.com \
--cc=sct@redhat.com \
--cc=viro@math.psu.edu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).