Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!mcvax!inria!lri!rd
From: rd@lri.UUCP (Roland Dirlewanger)
Newsgroups: comp.unix.wizards,comp.dcom.lans
Subject: TCP/IP between 4.2 BSD and non-4.2 BSD systems
Message-ID: <125@lri.UUCP>
Date: Tue, 2-Dec-86 17:22:19 EST
Article-I.D.: lri.125
Posted: Tue Dec  2 17:22:19 1986
Date-Received: Thu, 4-Dec-86 02:38:50 EST
Organization: LRI, Orsay. France
Lines: 58
Xref: mnetor comp.unix.wizards:236 comp.dcom.lans:51


Hi folks,

Hope someone on the net could help me to solve a problem on our LAN.

Here's the story:

We have actually three 4.2 BSD machines:
	- two ORIONs, build by High Level Hardware (Oxford, GB)
	- one VAX 11/750, with an Interlan NI1010 Ethernet Board

All three are connect on a ISOLAN fan-out (something like a DELNI),
and networking works fine. Once, we had a SUN 2 also connected to the net,
and had no problem with it neither.

After a while, we connected the box on a thick Ethernet Cable, and put
three workstations on this cable :
	- one Explorer Lisp-Machine (Texas Instrument)
	- two Hewlett-Packard HP9000/320 (running hp-ux, close to Sys V,
	  with a fast file system a la BSD 4.2)

Now:
	- telnet and rlogin work fine.
	- between two 4.2 BSD systems file transfers via rcp or ftp 
	  work perfectly.
	- between a non-4.2 system (hp or explorer) and a 4.2 system,
	  the file transfers work perfectly as well.
but:
	- from a 4.2 system to the non-4.2 systems the transfers
	  are incredibly slow: it takes 2 minutes to transfer
	  250 K bytes.

I first suspected the ISOLAN box, so I connected the Vax directly on
the cable. Nothing changed.

Then I tried with the trailers feature disabled:
	/etc/ifconfig il0 `hostname` up -trailers
Nothing changed.

So, I am now suspecting something's there might be a bug in the
implementation of TCP/IP on vanilla 4.2 BSD systems, which appears
only while transferring files from a 4.2 to a non-4.2 system.

Am I right ? 
If so, has anyone the fix for that bug.

If not, does anyone know what's going on ?

Many thanks in advance.

------------------
Roland Dirlewanger
Universite Paris-SUD
Laboratoire de Recherche en Informatique
Batiment 490
91405 Orsay Cedex

France