linux-kernel - Re: CFQ: async queue blocks the whole system

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1307927009.15392.102.camel@sli10-conroe>
Date:	Mon, 13 Jun 2011 09:03:29 +0800
From:	Shaohua Li <shaohua.li@...el.com>
To:	Vivek Goyal <vgoyal@...hat.com>
Cc:	Tao Ma <tm@....ma>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Jens Axboe <axboe@...nel.dk>
Subject: Re: CFQ: async queue blocks the whole system

On Fri, 2011-06-10 at 17:20 +0800, Vivek Goyal wrote:
> On Fri, Jun 10, 2011 at 11:02:11AM +0800, Tao Ma wrote:
> 
> [..]
> > > I don't think we can give a deadline for async request, because we
> > > still want to give sync high priority. We can give async some slices,
> > > so for a workload of small number of async requests and large number
> > > sync requests, we don't starve async too much. But for a workload with
> > > large number of sync/async requests, async will be starved for sure
> > > and we can't solve this in cfq.
> > OK, so if you guys thinks a 500 seconds wait is good for an async write
> > to complete, fine, then we have to switch to deadline.
> 
> I don't think that starving WRITES completely is a good idea. Especially
> given the fact that you were not able to dispatch WRITES for 500 seconds.
> This needs fixing.
> 
> Its not about giving hard deadline to WRITES, but making sure we don't
> starve them completely and they also make some progress.
Sure, I have no objection to avoid write starvation for a light write
workload. We definitely should do something for such workload. Your
patch is a good start.
But for a heavy write workload (for example, Ma's test), CFQ never can
completely avoid write starvation. In such workload, write queue has
only slow progress to handle requests.

Thanks,
Shaohua

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/