> I've suspected for a long time that wierd wedges I've seen on
> diskless Suns were due to a packet or two getting dropped during an
> NFS swap transaction, but I've been unable to produce this "hanging"
> effect on any of my diskless hp300s.  Though, it might be worth
> noting, my hp300s have much newer LANCE chips in them than most Sun 3
> systems...

That alone is not it, fwtw.

I have a Sun-3/260 that I once ran diskless for a solid week, doing a
full build of the world.  This was quite some time ago, well before the
cache-enable code for the -3/260 went in.  But recently, attempting to
run any Sun-3 diskless has failed pretty dismally for me.
Unfortunately, I wasn't actively tracking -current on that machine, so
I can't say when it started breaking....

How would a lost packet during a swap transaction cause trouble,
though?  Does swapping use different NFS code, or does the NFS code not
retry correctly in the face of lost packets, or something?  Perhaps
someone should tweak the ethernet driver to silently drop packets,
either input or output or both.  Something like that should be able to
identify a problem based on packet loss pdq.  If you crank it up to
lose 10% of the packets and it hangs almost instantly, but at .1% it
takes about 100 times as long, well....

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu