From owner-freebsd-hackers  Wed Jun 18 11:57:23 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id LAA03582
          for hackers-outgoing; Wed, 18 Jun 1997 11:57:23 -0700 (PDT)
Received: from phantasm.scl.ameslab.gov (phantasm.scl.ameslab.gov [147.155.142.74])
          by hub.freebsd.org (8.8.5/8.8.5) with SMTP id LAA03577
          for <hackers@freebsd.org>; Wed, 18 Jun 1997 11:57:19 -0700 (PDT)
Received: (from ccsanady@localhost) by phantasm.scl.ameslab.gov (8.6.12/8.6.12) id SAA05367; Wed, 18 Jun 1997 18:57:13 GMT
From: "Chris Csanady" <ccsanady@scl.ameslab.gov>
Message-Id: <9706181357.ZM5365@phantasm.scl.ameslab.gov>
Date: Wed, 18 Jun 1997 13:57:12 -0500
X-Mailer: Z-Mail (3.2.3 08feb96 MediaMail)
To: hackers@freebsd.org
Subject: Network concurrency problems!?
Cc: matt@3am-software.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Some background..  I have 2 ppro200 machines running 2.2.2 with 4 back
to back full duplex 100Mbps connections.  (and 2 10Mbps connections to
a real network.)  These are all SMC 10/100 cards, using Matt's 960603
drivers.  I am concurrently running NetPIPE over each of the links.  It
tests TCP, using a range of increasing block sizes.  (info at
http://www.scl.ameslab.gov/netpipe)

I don't know if this is a generic problem or driver specific, but when I
have 2-4 cards in a machine, I am only able to saturate one link with
larger transfer sizes. With 4 cards, at approx 6k, and 16k, two of the
pipes are completely starved.  The last runs about 10Mbps slower.  Also,
there appear to be hundreds of thousands of collisions on a couple of the
interfaces.  (This should be impossible, correct?)

Anyways, is this possibly a PCI int sharing problem, or some odd livelock
situation being encountered?

How exactly would I go about tracking this down?

Chris Csanady