We're running nfsd with 6 threads. However each thread appears to be using up to 20% of the cpu. I'm at 80% cpu busy for system only. No wonder stuff is slow.

We're also running an oracle db in case that matters. But it appears to be behaving nicely.

Thoughts on why nfsd would be using so much resource.

"80% CPU" does not mean that "the system is running so slow." The CPU is the one resource that you almost never wait for. It's a good thing if utilization is pegging 99.9% whenever there's anything to do.

Look elsewhere. Try hdparm and see if DMA is turned on. Look at top and vmstat 10 displays and see what the various processes in the system are waiting on.

I should have posted that the load average is between 10 and 20, and top shows nfsd at the top 6 or so cpu users. These are SCSI drives I'm dealing with.

vmstat shows plenty of memory available, no paging going on.
iostat looks okay.
sar -b shows average throughput. Here is a sample of the sar -b output:
tps rtps wtps bread/s bwrtn/s
10:30:03 AM 214.43 572.96 36.31 50.93 781.87
10:40:02 AM 168.26 488.50 34.88 49.47 765.73
10:50:02 AM 85.19 331.58 37.07 31.77 820.65
11:00:04 AM 86.73 346.14 35.53 84.62 748.65


This is running on an old kernel. RHES AS2.1 update5 (2.4.9 e49) in case that helps.

Thanks!!!!

Another tidbit. Oracle also runs on this machine. If I stop the oracle instance everything gets better.
So, I guess this is a database issue. Sure has me scratchin the ole noggin.