-----Original Message-----
From: Seth Kurtzberg [mailto:seth@cql.com]
Sent: Wednesday, May 15, 2002 7:55 PM
To: Stuart Brooks
Subject: Re: FW: Problem with rtk0 cards / driver


I would also very much like to solve this problem.  The same chip set is
showing up on SMC network cards, and also on some Pentium 4 mother boards
that have built in network hardware.

Unfortunately, the problem is not trivial.  It seems to occur on all
operating systems; perhaps I shouldn't say "all"; I've documented it on
NetBSD, Linux, and Windows 2000 Server.  Usually it isn't noticed on
Windows,
since people are used to rebooting the system occasionally for unknown
reasons, but it definitely occurs there.  I talked to an engineer at a
company building an embedded board using Linux who was also experiencing the
problem.  I had hoped that they would find the solution for me, but they
disappeared instead.  (I knew they had a lousy product, but couldn't they
have disappeared AFTER they found the Realtek problem?)

The key here will be to find a reliable way to recover when the problem
occurs.  It is possible to detect it.  For a while I was attempting to
recover by detecting the problem and reinitializing the network subsystem,
but unfortunately this doesn't always work.  Sometimes, even after
resetting,
you can ping out but other hosts cannot ping in.  I don't know whether this
is related to the RealTek problem per se or is perhaps related to higher
level network software in NetBSD.  I haven't had a chance to see/test this
behavior in Linux.

So, I'm open to suggestions, but I'm fresh out of good ideas at the moment.

On Wednesday 15 May 2002 08:46 am, you wrote:
> Sorry to hassle about this but I've really been struggling with this. Does
> anyone have any ideas aside from replacing the nics with different models?
> Reliability is quite an important issue, and unfortunately so is supply as
> there is not a huge range of options here in South Africa.
>
> Any ideas would be *much* appreciated,
>  Stuart
>
>
> ===============
>
> I have had a few issues with the Realtek 8139 cards I am using, and I am
> not sure whether it is a hardware or driver problem. My setup is as
follows
> :
>
> I have 3 i386 machines running NetBSD 1.5.1 and they are all sitting on
the
> same hub.
>
>         ------------------------
>
>         |   ftp    |           |
>
>         A <------> B           C
>        38.1       38.2       38.4
>
> If I do a large (500meg+) ftp from A to B, while at the same time doing an
> ftp from B to A, everything is happy. (each machine can ping every other
> one)
>
>         ------------------------
>
>         A -------> B ------->  C
>             ftp        ftp
>
> However, if I am doing a large ftp from A to B, and then start an ftp from
> B to C, I begin to see problems. Usually the network interface on A (or B)
> will appear to stall after a little while (usually between 10 seconds and
a
> minute), and it will be unable to ping any of the other machines. The ping
> fails with the message :
> "ping: sendto: No buffer space available"
>
> If I do an 'ifconfig rtk0 x.x.x.x', the interface wakes up, and I can ping
> etc. etc. However if the ftps are still going on it will stall again after
> a little while.
>
> It is not always the same box which stalls, but at least one of them does
> *every* time. The fact that it is not always the same machine which has
> problems would suggest that it is not a flakey card??
>
> Has anyone seen these kinds of problems before, any ideas what could be
> causing them, and any solutions?? Would it be a good idea to replace *all*
> of the network cards and give it another bash? How about the hub, could
> that possibly be the problem?
>
> Any help would be much appreciated...
>  Stuart

--
Seth Kurtzberg
MIS Corp.
Office:  (480) 661-1849
Fax: (480) 614-8909
email:  seth@cql.com
pager:  888-605-9296 or email 6059296@skytel.com