From owner-freebsd-hackers  Fri Nov  3 10:59:29 1995
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id KAA23488
          for hackers-outgoing; Fri, 3 Nov 1995 10:59:29 -0800
Received: from netcom22.netcom.com (bakul@netcom22.netcom.com [192.100.81.136])
          by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id KAA23466
          for <hackers@freebsd.org>; Fri, 3 Nov 1995 10:59:26 -0800
Received: from localhost by netcom22.netcom.com (8.6.12/Netcom)
	id KAA29234; Fri, 3 Nov 1995 10:59:04 -0800
Message-Id: <199511031859.KAA29234@netcom22.netcom.com>
To: hackers@freebsd.org
Subject: Network oddity between two FreeBSD machines.
Date: Fri, 03 Nov 95 10:59:03 -0800
From: Bakul Shah <bakul@netcom.com>
Sender: owner-hackers@freebsd.org
Precedence: bulk

I am running into a strange network problem.  Perhaps
someone else here has seen the same problem and solved it?

Here is the net setup:

	....----+--------------+----....  ethernet-A
		|              |
		|              |
	       one            two
		|              |
		|              |
	....----+--------------+----....  ethernet-B

One and two are FreeBSD machines, both running routed -q.
One has link address of one-A on net-A and one-B on net-B.
Two has link address of two-A on net-A and two-B on net-B.

Net-A has been around for a while and a number of other
machines are on it, including the nameserver.  One and two
have their primary net interface to net-A.

Pinging two-A from one works fine (and takes about 0.7 ms).
Pinging one-A from two also works (and takes about 0.7 ms).

Pinging two-B from one or pinging one-B from one is the
weird case: the ping time starts out high (anywhere from 10
ms to 1+ second) and with every ping the response time goes
down by 10 ms until it reaches 10 ms at which ping response
time bumps up to 1000 ms and the pattern repeats: 1000, 990,
980, ... 20, 10, 1000, 990 ...

This symptom does not change even when I remove all other
machines from net-B.  Ditto with using dotted inet address
instead of names and pinging to only directly connected nets
(ping -n -r).

I spent some time tracking this but gave up as more pressing
things had to be done but I am curious as to why this
happens.  That 10ms reduction on successive pings sounds like
some internal timer going off.

Just in case it matters, both machines are P100 Gateway PCI
machines, and are using 3COM 509 cards for both interfaces,
and running recent SNAPs but not the latest one.  Net-A has
Linux/BSDi/Solaris/IRIX etc. as well as win-NT and win-95
machines.

Anyone seen anything like this?  Yeah, I know I can ask on
comp.protocols.tcp-ip but thought I'd ask here first.
Thanks for any insight!

--bakul