I have just upgraded an i386 from 3.9 to 4.0. It's an MX mail server
that writes emails to another PC via NFS.
The delivery of the email via NFS is now VERY slow.
I noticed that when more then one precess try to access the filesystem
via NFS it is very slow. Even a simple "ls" of a small directory takes a
few seconds. If I kill all processes, then accessing that directory is
again very fast.
Please note that CPU usage is usually low (a great percentage of "idle"
CPU) but Load Averages are very high (20 and more) due to the many
processes trying to write via NFS.
To increase NFS throughput I had increased (many months ago) the number
of nfsio to 20 in sysctl.conf (vfs.nfs.iothreads=20). Here is their
current status:
10109 ?? IKL 0:00.03 (nfsio)
23992 ?? DKL 0:01.20 (nfsio)
15951 ?? IKL 0:00.57 (nfsio)
16583 ?? IKL 0:00.28 (nfsio)
15549 ?? IKL 0:00.13 (nfsio)
1027 ?? IKL 0:00.09 (nfsio)
10957 ?? IKL 0:00.07 (nfsio)
25036 ?? IKL 0:00.06 (nfsio)
12032 ?? IKL 0:00.05 (nfsio)
6440 ?? IKL 0:00.04 (nfsio)
17435 ?? IKL 0:00.03 (nfsio)
8590 ?? IKL 0:00.02 (nfsio)
15924 ?? IKL 0:00.02 (nfsio)
24621 ?? IKL 0:00.02 (nfsio)
7798 ?? IKL 0:00.02 (nfsio)
26897 ?? IKL 0:00.02 (nfsio)
26366 ?? IKL 0:00.02 (nfsio)
15218 ?? IKL 0:00.02 (nfsio)
24631 ?? IKL 0:00.02 (nfsio)
31798 ?? IKL 0:00.02 (nfsio)
Please note that I upgraded only the "client" PC, no changes on the NFS
server (an OpenBSD i386 3.9).
There have been changes in 4.0 that can explain this behavior?
Is there something I can do to solve it?
Thanks.
--
___________________________________________________
__
|- [EMAIL PROTECTED]
|ederico Giannici http://www.neomedia.it
___________________________________________________
