I swear, not all my questions are NFS related. Just the weirdest ones.

I have a VxWorks client that connects to a RHEL5 NFS server over UDP.
The VxWorks device needs to copy a 6MB file into RAM to make something
happen.

Anyways: connecting to the RHEL5 NFS server over TCP from another
Linux machine, this file transfer takes a few seconds. Sadly, UDP is
my only option because that's what was implemented in the kernel. Over
UDP (both from the VxWorks client and a linux client mounting with
just '-o udp'), I see unpredictable results:

- Sometimes it will take a few seconds.
- Sometimes it will take 5ish minutes.

I tried setting the rsize/wsize on the Linux machine to slightly less
than the MTU (<1500) and things worked at a reasonable rate again,
consistently. Sadly, VxWorks has no facility to set rsize/wsize; the
socket buffer is coded in the kernel at 10000 bytes.

When things are acting slow, tcpdump on the RHEL5 NFS server shows
bursty traffic being sent - chunks of packets being sent, then a few
seconds pause, then some more chunks, then pause. When thing are
acting fast, those seconds pause are no longer seconds of pause.

Is there a server side setting I'm overlooking? Lots of googling says
no, but maybe you have seen something I have not.
Is there a network congestion thing I'm not taking into account? Not
entirely impossible -- I don't have any visibility into the switches
I'm using. The network team says they're not doing any prioritization
of udp over tcp, which I would not expect or believe but I figured was
worth asking.

What else should I look at? Because, I'm lost.

A million thanks in advance...

- Pat
_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to