[OpenAFS] Slow file writes

Burcin Erocal burcin@math.bilgi.edu.tr
Fri, 24 May 2002 00:22:15 +0300

On Thu, 23 May 2002 02:08:17 -0400 (EDT)
Mitch Collinsworth <mitch@ccmr.cornell.edu> wrote:

> We're seeing some strange behavior on some newly installed afs clients.
> The symptom is that writing files into afs on these systems is very
> slow.  These systems are Redhat 7.2, kernel 2.4.9-31smp, OpenAFS 1.2.3
> client.  Our file and database servers are Redhat 6.2, OpenAFS 1.2.2.
> The odd thing in all this is that the problem systems are all on the
> same subnet.  Other clients with same software versions on other subnets
> are fine.  This of course makes me suspect this might be a network-related
> issue, but so far that hasn't been provable.  Non-afs file transfers on
> that subnet are fast, and there are no other observed network-related
> problems there.  The network path from there to our fileservers has no
> links slower than switched 100 Mbps.
> Anybody seen anything like this before?

How is your network set up?

I have seen that clients behind an OpenBSD NAT box have really slow access to AFS. Clients on a different subnet, behind a CheckPoint FW1 doing NAT, have no problems.

I have also found that if I tell the OpenBSD bridge that connects the AFS server to the outside world, to scrub all the packets, then all the connections slow down.

This has something to do with fragmented packets, I don't know if anything can be done on the AFS side though.

Can anyone help?