When I was investigating this previously, I had determined that all of 
the
following were factors conducive to causing the problem:

- Low system memory. My main system has only 128 MB RAM, and usually 
runs
   about 60 MB into swap.
- Large SU/chunk sizes.
- softdep.
- More columns in the RAID5 set.
- swap on RAID5.

Increasing uvmexp.reserve_kernel works if it is increased to the point
that is can satisfy RAIDframe RAID5 requests. In all cases I've looked 
at
on my system, uvmexp.free has been <= 4, the "raid0" RAIDframe kernel
thread is wedged trying to allocate memory, and there have been several
hundred write requests in raid0's queue, the majority of the requests 
are
uvm/ubc dirty buffer flushes.

I never did find out how to work out which allocation the RAIDframe 
kernel
thread was hung on. And I still can't fully fathom why increasing
vm.nkmempages can help - I'm guessing it may have a positive effect on
systems with a larger memory footprint than mine.

I'm to the point of contemplating setting up a test system with a few
disks, either that or trying to reproduce this with RAID5 on vnd. I 
don't
particularly wish to break my main system.

Cheers,
--
Paul Ripke
Unix/OpenVMS/TSM/DBA
I love deadlines. I like the whooshing sound they make as they fly by.
-- Douglas Adams