Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!zaphod.mps.ohio-state.edu!usc!snorkelwacker!apple!rutgers!mephisto!udel!new
From: new@udel.EDU (Darren New)
Newsgroups: comp.sys.amiga.hardware
Subject: Re: Networking options
Message-ID: <18863@estelle.udel.EDU>
Date: 8 May 90 21:00:06 GMT
References: <1696@mindlink.UUCP>
Reply-To: new@ee.udel.edu (Darren New)
Distribution: na
Organization: University of Delaware
Lines: 27

In article <1696@mindlink.UUCP> a127@mindlink.UUCP (Dave Norman) writes:
>CSMA / CD nets have a non linear degradation with increased nodes as the
>likelihood of a "collision" of two nodes trying to talk simultaneously increase
>exponentially with number of nodes and network traffic.  So the greater
>bandwidth of ethernet is hindered by its collision detect everybody try again
>algorithims.  

This is starting to drift from purely Amiga topics, but...
If you look at the maximum number of computers on an Ethernet (based
on max cable length and min cable between units) and the max size of
a packet on ethernet, you find that the number of collisions on
an ethernet are extremely small. Even on busy nets (like many dozen
diskless workstations) the number of collisions is less than 10%.
(Note that that is an empirical measurement from a friend who manages
the network for a large PA bank.  I believe him.)
Yes, the increase of collisions is exponential, but for a 10Mbit
ethernet to be worse that a 2.5Mbit token ring, you would have to have
3/4 of all packets cause collisions. Look at how much bandwidth you
would be requesting to cause 75% collisions and you will see that
it is more than 2.5 Mbits.

The real advantage of a token ring is bounded delay.  That is, if you
are running you Amiga animation over a network in real-time, waiting
a tenth of a second for a particularly bad random timing of collision
resolution can cause quite a noticable glitch. Token rings don't have
this problem.
			       -- Darren