DragonFly kernel List (threaded) for 2004-04
Re: Strange NFS performance
Have you been able to rule out hardware problems?
I'm thinking specifically of the cabling -- I've encountered far too
many questionable CAT-5 cables which have either been improperly
constructed (e.g., 3&4/5&6 paired instead of 3&6/4&5) or had one too
many chairs roll over them. Far less commonly, our switches in the
office also have a tendency to lose a port or forget their duplex
settings at times.
Richard Nyberg wrote:
Thanks for all response guys!
I've done some testing using kernels with last commit
i386/i386/busdma_machdep.c rev 1.9. I still have the same problem with nfs
but it doesn't seem to be nfs that's broken, but rather some network code.
This time I transfered data in one direction from a file on one machine
to /dev/null on the other using netcat. I did this with both tcp and udp
and in every permutation of source and destination machine. I monitored
the traffic with 'netstat 1'. I repeated the tests to be sure. The results
were very stable.
Comp1: DragonFly server (500Mhz PIII, 384MB RAM, if_fxp)
Comp2: FreeBSD 4.8 client (750Mhz AMD, 256MB RAM, if_vr)
Comp3: DragonFly client (1700XP AMD, 512MB RAM, if_vr)
This is the two tests that didn't perform as they should:
Comp1 -> Comp3 tcp: very, very bad and erratic transfer rate
Comp2 -> Comp3 tcp: bad and erratic tranfer rate
All other permutations showed good, stable tranfer rates.
I even tested Comp3 with fbsd5.2 without trouble.
Hopefully this means something to someone :)