"Thomas Hertz" writes:
> > > My system presently has 512Mb of memory, so I guess this 
> > has something 
> > > to do with the added disk io that comes with swapping on the raid 
> > > array.
> > 
> > Oh!!  Are you swapping to the RAID-5 array??  There are known 
> > problems with doing that....
> 
> Actually, I'm not swapping to raid-5 even though I would like to swap to
> a raid-10. But I used to swap on my raid-5, but I cannot say if that
> made things worse or not.

Swap to RAID 5 will definitely cause problems.  (I've been able to 
easily reproduce that problem..)

> I've recompiled with KMEMSTATS and really don't know what to look for.
> During normal use (no parity rewrite, no reconstruction) vmstat -m tells
> me that raidframe is using up to 515k memory, which sounds reasonable. I
> then tried failing a component (raidctl -f), and noticed that suddenly
> RAIDframe started using one 524288 byte size bucket in addition to the
> 64k buckets during normal operation. Also, RAIDframe uses up well over
> 30 megs of kernel memory (instead of the earlier 515k).

Yikes!!!!! 

> But then again, I don't know what's reasonable. :)

It sounds quite high...  but the more components, the more 
"OUTSTANDING" IO's allowed, and the larger the stripe width, the more 
memory it will use... 

> I'm gonna try to see what happens if I remove some of the NICs sometime
> tonight. Could it be that the NIC driver (tlp) is behaving badly in some
> way?

I suppose it could be, but I doubt it.... 

Later...

Greg Oster