[oclug] kswapd problems?

[Martin Hicks]

On Wed, 2002-08-28 at 19:18, Milan Budimirovic wrote:
> I have encountered a strange problem with the kernel swap daemon. I have
> a twin PIII-866 server with 3G of RAM. Because I do a lot of keyword
> indexing, I added two swap partitions which totalled a little less than
> 1G (I didn't want to butt up against the 4G barrier). I found that as my
> indexer soaked up more and more RAM, the server crawled to a
> near-standstill. So, I decided, "Hmm, I think I'll just tell the indexer
> to use less RAM and remove the swap partitions. I don't reaaly need the
> extra Gig." BAAAAAD idea. kswapd began to eat clock cycles like there
> was no tomorrow. It was using 99% of the CPU when it wasn't supposed to
> be running at all. What would cause that?
> 
> I am using a 2.4.9 kernel with SMP support. I seem to recall that there
> were some VM issues with the 2.4.x kernels. Is this something that a
> newer kernel is likely to fix? I've built a 2.4.18 kernel but haven't
> installed it yet.


VM in 2.4.9 is broken.  You _must_ have a pile of swap (I believe it's
twice the amount of physical RAM).  Search through the linux-kernel
mailing list archives for exact details.

I recall that the 2x physical RAM size is basically a requirement, not a
suggestion.

mh

-- 
Martin Hicks  ||  mort at bork.org  || PGP/GnuPG: 0x4C7F2BEE
plato up 14 days, 14:00,  4 users,  load average: 0.16, 0.30, 0.26
Beer: So much more than just a breakfast drink.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://tux.oclug.on.ca/pipermail/oclug/attachments/20020829/683e1e30/attachment.bin