Jeff Sturm wrote:
>
> Alvin Starr wrote:
> > Linux has a bit of a Feature;) in that when there is a lot of disk buffers
> > that need to be written out to disk, Linux will write them all out at once
> > and take some time to do it. While this is happning things will generaly
> > grind to a halt.
>
> Yep. Linux/x86 exhibits the same symptom, on a 32MB system no less...
> part of the problem I'd bet is that Paul is reading from the /dev/zero
> device, which if I'm not mistaken is used to allocate empty pages from
> swap. So you've got the double whammy of disk buffer and swap activity
> at once.
Hmmm... I'll have to boot a uniprocessor kernel on the dual Celeron
I've
got at home and see if it does the same thing. That would pretty much
clinch the diagnosis, I think.
> I don't see it when copying plain files.
We don't see it with small files. My team-mate noticed it while catting
files together to try to make a file bigger than 4Gb.
> I'd be a little surprised if a
> real application were seriously affected.
Me too. "Real" applications like Netscape and xterm run just fine.
And I imagine even the big number-crunching CFD codes the engineers
downstairs want to run won't have a problem, since they won't be
doing sustained huge file I/O.
Thanks to all of you for your thoughts. I think I know what to tell
the bosses now to reassure them.
Paul Allen
-- Paul L. Allen | voice: (425) 865-3297 fax: (425) 865-2964 Unix Technical Support | paul.l.allen@boeing.com Boeing Phantom Works Math & Computing Technology Site Operations, POB 3707 M/S 7L-68, Seattle, WA 98124-2207-- To unsubscribe: send e-mail to axp-list-request@redhat.com with 'unsubscribe' as the subject. Do not send it to axp-list@redhat.com
This archive was generated by hypermail version 2a22 on Tue Aug 1 04:53:12 2000 PDT
Send any problems or questions about this archive to webmaster@alphalinux.org.