> > Nice discussion fo the issue. One comment I'd add, though: since running out
> > of swap is catastrophic, it's better to over- than underallocate it.
> 
> And what are the specific castrophie modes?

 - inability to fork new processes
 - running processes get killed when they touch pages which force the
kernel to allocate memory for them (e.g., touching zero-filled pages,
or writing to copy-on-write pages).

The processes which get killed are not necessarily the ones which
*caused* the swap shortage, so this is likely to affect other
long-running, large processes, like your X server, emacs, or your
favorite long-running application or server, ..

 - kernel failures/deadlocks (older versions of the NetBSD kernel have
not dealt well with running out of physical memory; several of these
problems were fixed in the 1.4 timeframe but no doubt some others
remain..).

					- Bill