Subject: Re: unified buffers and responsibility
To: Wojciech Puchar <wojtek@chylonia.3miasto.net>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: tech-kern
Date: 06/12/2002 22:55:56
On Wed, Jun 12, 2002 at 08:31:19PM +0000, Wojciech Puchar wrote:
> > > However I did notice a strange thing:
> > > for various reasons on both systems I tested,, the copy goes from/to
> > > a different disk than the system disk. While I do the copy,
> >
> > OK, I newfs an old linux partition (which doesn't boot any more anyway).
> > doing a cp on the same disk as the system disk freeze X until the
> > cp is done.  Same result if I use dd instead of cp.
> 
> yes EXACTLY that!
> maybe not complete freeze but up to 10s delays. on your machine with i
> think faster IDE bus it may be worse.
> 
> 
> > Either reading or writing a large file cause this problem (dd if=file
> > of=/dev/null, or dd if=/dev/zero of=file)
> 
> 
> i often do dd if=/dev/rcd0d of=file bs=2k to take CD image
> 
> in 1.5.2 i was normally using machine while doing it. now i can't.

I still can do that, but it's because my IDE disk is fast, I guess.
disk is only 10% busy.

> 
> >
> > Has anyone idea of what is causing this ?
> > I suspect it could be disksort, which cause a single request to one end of
> > the disk be delayed to group dozen of requests on the other end of the disk.
> 
> if so - apps that doesn't use disk at all for sure (like text editor)
> should work fine. but it isn't - delays are small but noticable (up to 1
> second).

it looks like file activity is still pushing out some data from RAM,
even if it should not.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--