MBUF Cluster Network freeze problem

[bob]

Hi Phil,
No luck yet I'm afraid. There are two scenarios that we are suspecting as
possible triggers for the condition that starts the cluster
allocation/deallocation problem.
1) The application layer may be badly behaved with regard to the sequence of
calls to the TCP/IP stack. We certainly have seen the stack (not the
application) breaking permanently very quickly when the wrong, or out of
order calls are made. Arguably, some of the sequence of calls might not be
"wrong" but the Stack doesn't like them that way. On occasion it has seemed
fairly sensitive to breaking by an unsympathetic application layer. Not
actually correlated this to running out of MBUF clusters though.
2) A busy network (i.e. after a burst of flood pings) - I had a telnet
session activated to the BSP and was able to monitor the Networks stats. I
saw the free cluster count decrease by a couple every time I dumped the
stats. Eventually, when none were free it stopped working (not
surprisingly). Interestingly there were still hundreds of free MBUF's. I
don't know enough about the BSD/TCP stack to understand the relevance of
this; does anyone know for example, when is it supposed to allocate a new
cluster, and when is it supposed to free one?

Sorry to be quite vague, especially with regard to scenario (1). It was a
while ago when we discovered this and we just re-ordered our code at the
time. Since you are using the more powerful PPC BSP, I suspect its not our
68360 CPU temporary overload situation causing the cluster problem - which
is a relief.
I have not tried a new snapshot yet - it may take me a while to get it
together as I am only part-time on this, please let me know if you get there
first.
Thanks for your interest.
Bob Wisdom

-----Original Message-----
From: phil at zetron.com [mailto:phil at zetron.com]On Behalf Of Phil Torre
Sent: 01 December 2000 18:41
To: bobwis at ascweb.co.uk
Subject: Re: MBUF Cluster Network freeze problem

Hello Bob,

I read your message to the RTEMS list about your problem running
out of mbuf clusters.  Just this morning, I came in to discover
that my MPC860 system was doing the same thing.  I just had it
plugged into our office Ethernet serving some documents via the
GoAhead server; there's plenty of traffic on the LAN, but to the
best of my knowledge there were no flood pings or other stresses
as you described.  I'm also running beta3a.

You mentioned that you might try newer snapshots to see if the
problem persists.  Have you had any luck?

Thanks,
-Phil

--

=====================================================================
Phil Torre                               phone: 425-820-6363 x234
Design Engineer                          email: ptorre at zetron.com
Radio System Accessories Group             fax: 425-820-7031
Zetron, Inc.                               web: http://www.zetron.com