Date: Wed, 14 Mar 2001 15:43:23 -0800 From: Alfred Perlstein <bright@wintelcom.net> To: chat@freebsd.org Subject: Re: Disk I/O problem in 4.3-BETA Message-ID: <20010314154323.O29888@fw.wintelcom.net> In-Reply-To: <Pine.LNX.4.10.10103150931040.3321-100000@blow.research.canon.com.au>; from iain@research.canon.com.au on Thu, Mar 15, 2001 at 09:54:51AM %2B1100 References: <20010312151315.F18351@fw.wintelcom.net> <Pine.LNX.4.10.10103150931040.3321-100000@blow.research.canon.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
* Iain Templeton <iain@research.canon.com.au> [010314 14:54] wrote:
> On Mon, 12 Mar 2001, Alfred Perlstein wrote:
>
> [redirected to -chat, hopefully]
>
> > > It's actually worse. Someone, I forget who, ran some tests with
> > > write-caching turned on and found that the IDE drive could hold a
> > > pending write in its cache 'forever', even in the face of other writes,
> > > as long as there was other disk activity going on. So we aren't just
> > > talking about issuing I/O's out of order, we are talking about issuing
> > > a sequence of writes and having some of them simply not ever commiting
> > > to disk (not for a long, long time) in a heavily loaded environment.
> > > That's bad news.
> >
> > Someone leaked the Linux austrailian elevator algorithm to the disk
> > manufacturers?
> >
> As an elevator user in Australia I'm intrigued by this comment?
Just picking.. :)
>
> I know our elevators at work can play up ("No I said up to the 5th, not
> down to B1!").
It's a common starvation problem.
Imagine an elevator that always goes to the nearest requested floor
instead of making almost complete up then down sweeps.
At quit time there's a good chance you'll never hit the lobby
because the elevator keeps going to the nearest floor which is
unlikely to be the lobby once it gets high up enough.
Matt explained that Linux had this "feature" to reduce disk seeking
for a short period, finally someone with algorithm experience stood
up and smacked enough people around such that it got changed to
the proper elevator algorithm.
--
-Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org]
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-chat" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20010314154323.O29888>