"Andrés A. Nogueiras Meléndez"
I've noticed an extremely slow performance of NFS when a SuSE Linux 8.1 client (2.4.19-4GB) reads from a Solaris 7 (5.7 Generic_106541-18) server via NFS. Can anybody confirm the problem or is it specific to my configuration?
try to put in your /etc/auto.home
user -rw,rsize=32768,wsize=32768 machine:/export/dir/user ^^^^^^^^^^^^^^^^^^^^^^^^^^^
These options increases the buffer to the size of Solaris buffers.
I've changed the mount parameters to rw, rsize=32768, wsize=32768 but the performance hasn't improved. Actually, it started to fluctuate, sometimes I got the transfer speed 320 KiB/s, sometimes 780 KiB/s. I shortly looked into the tcpdump file I took. Most of the time the packets were sent with the maximum speed but often when there was a multicast (ALL-ROUTERS.MCAST.NET record) the NFS transfer was interrupted for approximately 1s (sometimes more). There were about 20 multicasts in the dump so they might account for the extremely long transfer time (about 30s instead of 1s for 10 MiB file). To be sure, a time analysis of the tcpdump file must be performed. Are there any tools? Anyway, it would be helpful to know that someone has a well performing NFS with a SuSE 8.1 client and a Solaris server (i.e. transfer speeds at least 5 MiB/s.)
Also watch your traffic in the net and, the last but no the least, check your wires. Sometimes, with high traffic deffective wires show up
rcp is not affected by those multicasts and I'm almost sure NFS performance was quite good in SuSE 7.3 (2.4.16 kernel). Thus I think the problem is more software than hardware specific. -- Alexandr.Malusek@imv.liu.se