On Wed, Dec 24, 2003 at 09:31:58AM +1100, Norman Gaywood wrote:
> I just upgraded from RH8.0 to FC1 on a Dell 6600. 4 processors, 4Gig ram,
> heaps of swap enabled. All current yum updates applied. tape drive is a
> PV 122T (200Gig tapes).
>
> What used to take dump (RH8.0) about an hour to backup 80 Gig to tape now
> looks like will take about 12 hours. I killed the dump. The CPU load every
> now and then will go upto around 30 with most of that time in system time.
>
> So something is broken. Is it dump or the writing to a tape drive? I'll
> do some experiments later and see if I can figure out more.
>
> Anyone seen anything like this.
I just worked it out (partly). It's a VM problem in the linux kernel!
If I do this before I start dump:
perl -e '$i=2000;while ($i--){ $a[$i]="x"x1_000_000; }'
I briefly creates a about 1.8Gig process and clears some room in the
kernel disk cache. A large amount of memory then becomes free. top will
show you the details.
With the cleared cache, dump runs quickly like it used to in RH8.0
Sigh. I've seen these sort of problems before in 2.4.18 around this time
last year:
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=79257
I don't know if it's just dump that triggers this or any big disk copy. I
guess I'll get around to testing this soon.
After a beer.
--
Norman Gaywood, Systems Administrator
School of Mathematics, Statistics and Computer Science
University of New England, Armidale, NSW 2351, Australia
norm@xxxxxxxxxxxxxxxxx Phone: +61 (0)2 6773 2412
http://turing.une.edu.au/~norm Fax: +61 (0)2 6773 3312
Please avoid sending me Word or PowerPoint attachments.
See http://www.fsf.org/philosophy/no-word-attachments.html
