Xref: utzoo alt.sys.sun:4329 comp.dcom.lans:8452
Newsgroups: alt.sys.sun,comp.dcom.lans
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!caen!acc.flint.umich.edu!jal
From: jal@acc.flint.umich.edu (John Lauro)
Subject: Re: Cabletron Repeater Problems
Message-ID: <Aph#+3-@engin.umich.edu>
Date: Mon, 24 Jun 91 16:00:42 GMT
Organization: University of Michigan - Flint
References: <1991Jun20.203445.12130@unhd.unh.edu> <1991Jun21.212938.29822@usenet.ins.cwru.edu> <1991Jun24.000210.4759@qualcomm.com>
Lines: 49

In article <1991Jun24.000210.4759@qualcomm.com> antonio@qualcom.qualcomm.com (Franklin Antonio) writes:
>I just look at the total number of collisions reported for a port, and
>divide by the number of packets on that port.  We often have these numbers
>in the 10% to 15% range on Cabletron repeater ports which have nothing
>but a single Shiva FastPath on them.   
>
>
>>>>discovery.  We can reduce the number of collisions by choosing WHICH
>>>>ports we use for the Shiva FastPaths...
>
>>  It make perfect sense here.  If you put less devices on the THN-MIM port
>>the less collision percentage it will display on the PCOV.  You you have done
>>is using more ports on you THN-MIM at the same time, you have just distribute
>>the collision packet transmission per port.  The total collision percentage
>>of the THN-MIM bord still remain the same but the percentage per port will be
>>much smaller now - see the defination.
>
>Now, given that, which ports on the MMAC should i use?  Are all repeater
>ports created equal?  If i pull some cables from the front of the THN-MIMs
>and plug them in different ports -- this is _just_ a _permutation_ of ports!
>-- i would expect all the collision statistics to remain the same.
>(of course now reported on different port numbers, because i have moved
>the "interesting" traffic to different repeater ports)
>What we seem to have discovered is that all repeater ports are _not_ 
>created equal.  The worst choice seems to be to plug those cables from
>the FastPaths into ports all on one THN-MIM card.  Still one repeater
>port per FastPath, and nothing else on those ports.  The best choice seems
>to be to plug the cables from the FastPaths into ports chosen one per
>each THN-MIM card.  Still one repeater port per FastPath, and nothing else
>on those ports.  This doesn't make sense at all for any reasonable 
>functioning of the repeaters.
>
>No way will i believe that permuting the cables at the front panel of
>the repeater should change the performance of my network.

This is all speculation, and I probabbly don't know what I am talking
about...

I don't know how the fastpaths work, but if they advertise information
based on responding to broadcast then 15% collisions of an idle network
sounds reasonable.  Basically they all receive a packet, and then all
respond at the same time.  By changing the cable arragement just slightly,
you minutely alter the timing from port to port just enough to allow the
collision avoidance built into ethernet controllers to operate better.
I would expect (but do not know for sure) that the timing from port to
port on the same card is slightly smaller than from port of one card to
port of a second card.

   - John_Lauro@ub.cc.umich.edu