Path: utzoo!utgpu!news-server.csri.toronto.edu!csri.toronto.edu!mart
Newsgroups: comp.dcom.lans
From: mart@csri.toronto.edu (Mart Molle)
Subject: Re: Ethernet with higher data rates ?
Message-ID: <1991May7.104158.2923@jarvis.csri.toronto.edu>
References: <heibe.673604290@forst>
Distribution: comp
Date: 7 May 91 14:41:59 GMT
Lines: 47

heibe@forst.informatik.rwth-aachen.de (Bernd Heinrichs) writes:

>Perhaps this is not a realizable idea. But I think in some environments it
>would make sense to use the existing Ethernet installations running with
>higher data, e.g. up to 100 Mbit/s, instead of the expensive FDDI.
>My problem is, that I do not know, if it is physically possible to run 50-
>or 75-ohm coaxial cables at such high data rates. Of course the cable
>length must be shortened and the minimum packet length has to exceed
>64 bytes.

>I would be very pleased when som people can tell me, such data rates are
>realizable or not.

Increasing the data rate by a factor of 10 in Ethernet is not going to work
very well, because you cannot also make the signals travel 10 times as fast
along the wires.  The efficiency of CSMA/CD is a function of the ratio of
*time* spent transmitting data to the propagation delay.  Thus, increasing
the data rate without shortening the network or forcing the minimum packet
size to go up makes this ratio (and hence Ethernet efficiency) drop to 
unacceptable values.

However, there are a number of proposals for "speeding up" Ethernet that
end up doing what you want but via an approach that avoids the CSMA/CD
bottleneck.  Basically, the idea is to build the Etherent analog to a PBX:
"low speed" (almost) dedicated lines (in this case 10Mbps Ethernet, instead
of a 64Kbps digitised voice circuit) leading to a "switch port", with the 
internals of the switch using some proprietary very-high-speed internal
architecture to move data between ports.  The Kalpana EtherSwitch promises
to do this using an internal 32x32 non-blocking buffered switch that supports
store and forward with cut-through Real Soon Now, and SuperNet (nee Hubnet)
has a product that uses a 100 Mbps collision arbitration rooted tree
network to connect a bunch of "simulated" multiport transceivers (which
looks like a DELNI, say, to the hosts but connects to the SuperNet backbone
using the Hubnet protocol at 100Mbps).  I guess one could say that these
systems support "tunneling" of Ethernet frames between ports.  There is also
an issue of how to decide *where* to send a frame that originates from a
given port, which was discussed in this forum recently....


These designs give a total bandwidth comparable to FDDI (or even higher),
but limit the host-to-host bandwidth to Ethernet speeds.

Mart L. Molle
Computer Systems Research Institute
University of Toronto
Toronto Canada M5S 1A4
(416)978-4928