🚀 go-pugleaf

In article <1992Nov19.193651.21854@umr.edu>, rfranken@mcs213f.cs.umr.edu (Richard Brett Frankenberger) writes:
> In article <sige6ao@rhyolite.wpd.sgi.com> vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
> ...
> >-Ethernet is simpler and so more robust.  Any single ring scheme is a
> > royal pain compared to ethernet.  The hassles of source routing and
> > other token ring "features" must be experienced to be believed.
> 
> It's more robust?  Really?  What happens if a faulty transciever cable in
> ethernet is messing up the entire line?  Nothing.  On Token-Ring?  The
> workstation will remove itself from the ring.  I can bring down an ethernet a
>lat easier than I can bring down a token-ring.  And I can get a token-ring back
> up faster than an Ethernet.
> ...

When comparing modes of failure, you must compare apples-to-apples.
Faulty transceiver cables very rarely mess up entire ethernets.  They
rarely mess up more than the station using them.

How does a station on a single ring like token ring "remove itself" if
the cable between it and a neighbor is cut?  No, you can't bring in
hubs, because there are also hubs for ethernet with very robust
per-link fault detection and isolation stuff.  (The answer is that
the ring is broken and dead.)


> ...
> >-10MHz Ethernet is faster than 4Mb/s token ring.  A pair of 16Mb/s token
> > ring stations can be 60% faster than a pair of ethernet stations,
> > but mostly because of the underlying bandwidth.  At the (now) low
> > speeds of ethernet and token ring, there is no interesting effeciency
> > difference for stations (not hubs or switches) between packet sizes
> > of 1500 and 16K.
> 
> Really?  On a 4 MBPs token-ring, I ran some large file copies at
> approx 1.5K and
> 4K packet sizes.  The 4K was about 25-50% faster (I can't remember exactly).
> Of course, for smaller transactions, the difference is less, but still
> noticable.  (This test was done with a lightly loaded Novell file server with
> Packet Bursting (without packet bursting, the difference would be even more
> pronounced).

I don't know about Novell, but that is not true with a Real Protocol (tm).
Any decent machine can saturate an ethernet today with IP.  No decent
machine does less than 8Mbit/sec over ethernet with IP.  8Mbit/sec is
more than will fit in a 4Mb/s token ring.  In my book, 16Mbit/sec is a
puny speed at any packet size above 1KByte.  There are performance
hurdles with 16Mb TR, but as I understand them, they are related to
common silicon, and not the stations.


> ...
> >Ethernet is far more common and not significantly worse than token
> >ring.  Since token ring is not significantly better (it can be argued
> >to be worse since it is so much more fragile), token ring is dead in
> >the long run.  The universe does not tolerate competing, largely
> >equivalent standards forever; eventually the less common one disappears.
> >
> >Something that might hurry up the decades long death of token ring
> >might be higher speed LAN's.  The short term winner(s) in those wars
> >might have some effect on the numbers of new ethernet and token ring
> >networks.  I thinking of cable plants and hubs that can be used for
> >for both high and low speed links to the desktop.
>  
>Token-ring won't die.  Eventually, FDDI will be much more common than E-net and
> TR.  FDDI is token-ring over fiber-optic, basically.  Then token-ring concept
> is superior to the CSMA/CD concept used by Ethernet.  It is more robust,
> have lots of error checking and correction built in, and  it is designed to
> take heavy loads into account and share the bandwidth as fairly as
> possible (Ethernet's method is just let them collide and we'll randomly pick
> who gets to send).

Not so.  I am somewhat immodest about some FDDI stuff; to the best of
my knowledge, the TCP/IP/FDDI stations I work on are among the fastest
in the world.  No one will ever be any faster, at least on 125Mbaud FDDI.
However, FDDI will never, ever replace ethernets.  Like comparing a
Model T Ford and a Porshe, FDDI is about 10 times faster than ethernet,
but less reliable, more difficult to install, and more expensive.
There are a host of cold technical reasons why those facts will not
change.  Ever.  (Yes, I've heard of CDDI; check my email address.)

Ethernet will endure forever like RS-232, just as RS-232 endures
despite the improvements of the standards bodies for these last 25 or
30 years.  Ethernet is too cheap, easy to build, install, and run.

FDDI may out live IBM Token Ring depending on how ATM, FC, and HIPPI
develop, but that is small consolation.


>Bottom line,  You probably won't go wrong with either, but with token-ring, you
>won't have to worry about one user sticking a pin through the cable and downing
>the entire net.  (Not that that's a likely occurance, but it makes the point
> about token-ring's robustness).  Ethernet will likely be cheaper, however.

There are ways to kill an entire ethernet, but not really with a pin.
(e.g.  power off the 10baseT hub, cut the coax, get creative with "coax
drop cables", reprogram MAC addresses on PC cards)  Have you ever tried
to stick a pin into yellow hose or even thinnet?  Without a drill?

What if you stick a pin thru the wire connecting TR hubs?


Vernon Schryver,  vjs@sgi.com