I have written a special-purpose NFS server that uses NFS loopback
(i.e., its clients run on the same machine that the server itself runs
on). I am having a problem related to the number of biod daemons
running. If I run more than two biod's, the some of the UDP packets
received by the NFS server go into exponential retry behaviour. This
kills throughput.
I assume that this is due to buffer overflow somewhere in the kernel,
probably because of the way loopback UDP sockets are implemented? Is
there any way to fix this? Right now I have to decrease the number of
biod daemons to get decent throughput.
I am using SVR4/386. Thanks for your help.
Joe Buehler