> > Processes shouldnt be crashing
>
> I violently and vehemently agree to that. :-)
>
> I managed to dd an 8Mb block of zeroes to the NFS volume, although I got
> copious NFS not responding/ok messages. This took nearly half an hour,
> which amounts to about 5K/sec - somewhat disappointing. (Hopefully, I
> got the calculations right this time) With 1K blocks, I see a definite
> improvement, down to about 7 minutes (15-20K/s), but still too slow.
>
> Keeping an eye on the file on the Linux server, I see it is being
> written to in large blocks (64-200K at a time) at very long intervals.
> Another peculiarity seems to be that it is real fast up to, say, 1.5 to
> 2Mb, and then it bogs down, giving me the dreaded messages.
I have realy similar behaviour between two linuxes on very broken
ethernet. Trouble is, that one of computers is 386 with ne1000, and it
just drops packets if they come too fast (producing eth0: receiver
overrun). TCP can do about 200K/sec over this link, NFS is *much*
slower.
But this is probably not your case. This link is constantly slow. (Due
to retransmits.)
> There's also a long delay after dd has completed (output it's records
> in/out message), if that is any clue.
Pavel
-- I'm really pavel@atrey.karlin.mff.cuni.cz. Pavel Look at http://atrey.karlin.mff.cuni.cz/~pavel/ ;-).