Xref: utzoo alt.sys.sun:4281 comp.dcom.lans:8425
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!pacbell.com!ucsd!qualcom.qualcomm.com!qualcom.qualcomm.com!antonio
From: antonio@qualcom.qualcomm.com (Franklin Antonio)
Newsgroups: alt.sys.sun,comp.dcom.lans
Subject: Re: Cabletron Repeater Problems
Keywords: Cabletron MMAC Shiva FastPath Ethernet collision statistics bogus
Message-ID: <1991Jun20.075328.6540@qualcomm.com>
Date: 20 Jun 91 07:53:28 GMT
References: <1991Jun7.200603.9378@ns.network.com>
Sender: news@qualcomm.com
Organization: Qualcomm Inc., San Diego, CA
Lines: 72
Nntp-Posting-Host: qualcom.qualcomm.com

In article <1991Jun7.200603.9378@ns.network.com> sjs@eros.network.com (Steve Senum) writes:
>Our site has a number of Cabletron MMAC-8s, IRMs (and IRM2s), and THIN-MIMs.
>... have had a *lot* of problems with
>the Cabletron repeaters dropping packets. ...
>...
>I would be interested to hear from other sites that have seen a problem
>like this with Cabletron equipment.

Let me tell you about MY adventures with Cabletron repeaters!

We have a bunch of Cabletron thinwire repeaters: Several MMAC-8s filled with
THIN-MIM modules, IRM2s, and also about a dozen MR-9000C's. 

The first problem is that while the Cabletron repeaters have collision
detectors on each port, and gather collision statistics by port, these
circuits DO NOT succeed at informing me which Ethernet segment contains
the problem causing the collisions.  Many times I have walked into one
of our comm closets and seen EVERY collision light blinking (on MR-9000's)
or seen collision statistics accumulating for every port (on MMACs).
My response has always been to disconnect Ethernet segments from the
repeater(s) one by one, until the collisions stop.  What is the purpose
of these collision detectors on each port if they don't help you 
isolate problems to a particular port?  We've recently implemented
Remote-Lanview, so we can see all these statistics for all the repeaters
remotely.  This SOUNDS great, but since all the collision statistics are
unreliable, all this provides is a very fancy Windows program to display
random numbers.  I expect these statistics should be a TOOL i can use to
diagnose my network.  Do i ask too much?

Typically in each building we have an MMAC filled with THIN-MIM modules and
an IRM2.  In the same comm closet, we often have several Shiva FastPath
Ethernet-to-Appletalk gateway boxes which create the appletalk segments
within that building.  There are several remarkable things that occur
when a Shiva FastPath is connected to a Cabletron ethernet repeater.
Long ago, we used to run an ethernet cable from one port on the Cabletron
repeater to all the FastPaths.  (Nothing else on this cable, except, of
course, a terminator.)  The first problem is that the Cabletron repeater
reports a lot of collisions on this segment.  The really amazing symtom
is that we've seen the Cabletron repeater report lots of collisions on
the NEXT repeater port too!  In other words, say port 6 on the repeater
goes to the FastPaths.  Port 7 will report collisions.  This can happen
even if port 7 is connected to no other ethernet devices!  You can put
a terminator right on the BNC connector on the front of the box and it
will still report collisions on the port.  This provides even more evidence
that the Cabletron collision detection/reporting mechanism somewhat bogus
information.

We've had better luck when we use one repeater port for EACH Shiva
FastPath.  That's the way we've been running for a long time.  We still
get lots of collisions on these ports, sometimes 10% or 15%, but the
network still seems to work.  Just a few days ago we made yet another
discovery.  We can reduce the number of collisions by choosing WHICH
Cabletron ports we wire up to the Shiva FastPaths.  We used to take the
first THN-MIM card in the MMAC, and dedicate it to the FastPaths.
Now we find that we get many fewer collisions if we spread out the
ports which connect to FastPaths, putting no more than one FastPath
on each THN-MIM card, then using the other 7 ports on each THN-MIM for
anything else.  I'm not sure exactly what to conclude from this.  
Sure looks like this has uncovered some kind of timing problem in the
MMAC design, but that's just one possibility.  

Shiva and Cabletron have each indicated that they haven't seen this
before.  It can't be a simple piece of broken hardware, as all our
Shiva and Cabletron boxes do the same thing.  Must be a design flaw.

Anybody else use Cabletron and Shiva boxes together?
Anybody else got Cabletron collision bogosity?
Anybody out there using Cabletron's collision statistics AND finding
them meaningful?

Thanks.
Franklin Antonio