Markus W Kilbinger writes:
> >>>>> "Greg" == Greg Oster <oster@cs.usask.ca> writes:
> 
> Maybe my (quite non-specific) panic's with raidframe raid5 systems
> belong to this problem:
> 
>   http://mail-index.netbsd.org/current-users/2002/09/15/0000.html
> 
>   http://mail-index.netbsd.org/current-users/2003/02/28/0009.html
> 
> Playing with NMBCLUSTERS (up to 16384) didn't help really help.
> Retrospective starting use of raid5's swap on these machines would be
> a good correlation for the timepoint of the panic/hangs.

It's been quite a while since I've tested swapping to a RAID 5 set, but
I'm not aware of any particular problems with that... (I'd argue that if you 
have enough disks to make a RAID 5, then you have at least enough disks to 
make one RAID 1 set, and the write performance to that will be *much* better)

> Anything I could try/test for narrowing this problem?

The output of "vmstat -m" (you'll need "options KMEMSTATS" in your kernel 
config) from just before the panic/crash/hang would be useful.  You might also 
bump up the "NKMEMPAGES" value, and/or play with "RAIDOUTSTANDING" values to 
see how those change the crashing behaviour.

Later...

Greg Oster