Ethernet Vs. Token Ring

sui...@freenet.edmonton.ab.ca

unread,

Oct 19, 1997, 3:00:00 AM10/19/97

to

Hi :

Can I know the pos and con about Ethernet and Token Ring?

--
Best regards,

Martin Leese - OMG

unread,

Oct 20, 1997, 3:00:00 AM10/20/97

to

On 20 Oct 1997 13:04:15 GMT Jerry (je...@reillyplating.com) wrote:
>> > Can I know the pos and con about Ethernet and Token Ring?

>> In theory, token ring handles high volumes of traffic better than ethernet.
>> In practice, ethernet usually works just fine.
>> Token ring uses a token-passing method to handle node access to the network
>> media (ie. you can't talk unless you have the token). Tokens passing all
>> but eliminates collisions; a problem that plagues ethernet under heavy
>> load.

To expand on this a little: The problem with Token Ring is that
the token can get lost. To cope with this, a watcher on the ring
generates a new token if it has been missing a while. Unfortunately,
this sometimes results in two tokens on the ring. To cope with this,
a watcher destroys one token when it sees two. This partly explains
why token rings are more expensive. The other reason, and I suspect a
much more important one, is volume of sales. Ethernet is everywhere,
so is very cheap.

The problem with Ethernet, as Jerry explained, is that it handles
heavy loads badly. Heavy loads result in collisions which REDUCE
effective bandwidth just at the time you need it most. Under heavy
loads Token Ring degrades more gracefully.

Regards,
Martin
* NEW * E-mail: mle...@unb.ca * NEW *
WWW: http://www.omg.unb.ca/~mleese/
______________________________________________________________________
Want to know how Ambisonics can improve the sound of your LPs and CDs?
Read the Ambisonic Surround Sound FAQ. Version 2.7 now on my WWW page.

Jerry

unread,

Oct 20, 1997, 3:00:00 AM10/20/97

to

> Can I know the pos and con about Ethernet and Token Ring?

In theory, token ring handles high volumes of traffic better than ethernet.
In practice, ethernet usually works just fine.
Token ring uses a token-passing method to handle node access to the network
media (ie. you can't talk unless you have the token). Tokens passing all
but eliminates collisions; a problem that plagues ethernet under heavy
load.

Ethernet uses CSMA (carrier sense, multiple access), to access the network
media. Nodes listen on the wire for a break in the action before sending a
backoff signal to all other nodes. The problem is, when many nodes have
data to send at the same time (as would happen under heavy load), multiple
stations often start transmitting at the same time. A collision detection
mechanism tells the offending nodes to stop, and all colliding nodes wait a
random (but small) time before retrying. As the load increases near
saturation, the number of collisions increase, and very little productive
communication occurs.
With that said, ethernet is (usually at least) quite a bit cheaper.
Ethernet (over Twisted pair, as commonly implemented) uses multi-port hubs
to connect all nodes. Hubs a generally pretty cheap compared to the multi
access units (MAU's) that are required by token ring. It's been a while
since I've priced or looked at token ring hardware, so you may want to
check on that your self.
There are ways to use ethernet and minimize the drawbacks: you can use
switched ethernet (or a switched main hub) which sends packets only to the
network segement that contains the destination node.

As I said in the opening, I have found ethernet to be adequate for our
network.

Hope this helps.

JD Kratzer

unread,

Oct 20, 1997, 3:00:00 AM10/20/97

to

I am in a shop that is almost exclisively Token Ring. Ethernet prices are
about 1/3 the cost of Token Ring. I have had good success in both
environments. Ethernet is a faster speed, but that is going to change next
year when they come out with the 100mb token ring.
John Kratzer jkra...@cemexusa.com
Cemex USA jdkr...@aol.com
Houston, TX http://members.aol.com/jdkratzer/homepage.html

Jason Clifford

unread,

Oct 22, 1997, 3:00:00 AM10/22/97

to

JD Kratzer wrote:
>
> I am in a shop that is almost exclisively Token Ring. Ethernet prices are
> about 1/3 the cost of Token Ring. I have had good success in both
> environments. Ethernet is a faster speed, but that is going to change next
> year when they come out with the 100mb token ring.

Given that we already have Gigabyte Ethernet as well as 10MB and 100MB
versions and that, in line with recent history, 100MB Token Ring is
likely to be incredibly expensive I think it is unlikely that 100MB TR
will take off in a big way.

Only two or three of the TR manufacturers are even working on this with
the others playing a 'let's wait and see if there really is a market for
this' game.

Remember that until three or so months ago Token Ring installation were
being advised that the only reasonable upgrade path is 155MB ATM. This
may yet prove to be the case for those who do not want Ethernet.

I have used both TR and Ethernet a lot. Generally Ethernet works out not
only less expensive but also better value for money so long as one
designs the network properly. In a well designed network Ethernet will
not suffer performance problems until you exceed 70% of bandwidth - 70MB
on a 100MB Ethernet. Do you really need more than that? If so then ATM
is probably the most appropriate choice.

Jason Clifford
--
Please note that my return email address is jason.c...@domgen.com

As a service I provide analysis of viruses and poor grammar to senders
of unsolicited commercial e-mail at a rate of $500.00 per hour. Delivery
of said correspondence constitutes a request for the aforementioned
services at said price. Supply billing address.

ane...@bannerfoundry.com

unread,

Oct 22, 1997, 3:00:00 AM10/22/97

to

In regards to pricing of Token Rings - check out this press release from
Bay Networks http://www.baynetworks.com/News/Press/nz/nz970613.shtml They
are talking about a low-cost Token Ring workgroup switch. Are you
familiar with it?

In article <344E2E09...@127.0.0.1>,

-------------------==== Posted via Deja News ====-----------------------
http://www.dejanews.com/ Search, Read, Post to Usenet

Brad Knowles

unread,

Oct 22, 1997, 3:00:00 AM10/22/97

to

In article <62di49$k7e$2...@news.sas.ab.ca>, sui...@freenet.edmonton.ab.ca ()
wrote:

> Can I know the pos and con about Ethernet and Token Ring?

Ethernet is more common in most Unix/PC/TCP-IP environments, wherease
Token Ring is much more likely to be found in IBM Mainframe environments,
but Token Ring has it's advantages for Unix/PC/TCP-IP environments as
well.

Although "regular" ethernet has a theoretical limit of 10Mbps
throughput, the reality of it is that anything over 27% utilization
typically starts causing significant numbers of packets to be lost and
retransmitted, etc.... This is because Ethernet uses a method that
potentially allows more than one system to be speaking at a time, which
causes what's called a "collision", leading to both systems needing to
attempt to retransmit whatever it was they were trying to say and
hopefully not at the same time (again).

OTOH, Token Ring can pretty much use the full maximum theoretical
capacity (either 4Mbps or 16Mbps for newer implementations) because it
guarantees that only one system at a time will be "speaking".

The same basic technology used in Token Ring can also be found in
FDDI, which has a theoretical maximum throughput of 100Mbps (unless you're
doing something real fancy and probably *very* non-standard). Again, FDDI
can use virtually all of that theoretical maximum throughput, whereas Fast
Ethernet (also 100Mbps) can only effectively use about 27% of that
theoretical maximum (it uses the same technology as "regular" Ethernet,
just at ten times the speed).

However, very few machines can actually make full use of even Fast
Ethernet, so that might not be a problem for you. Nevertheless, under
those circumstances, you would probably still have more left over network
capacity with FDDI than Fast Ethernet, because you'd be using a larger
fraction of the practical maximum throughput.

Unfortunately, both Token Ring and FDDI tend to be quite a bit more
expensive than their Ethernet counterparts (not only in terms of Network
Interface Cards in the machines themselves, but also in terms of cabling,
concentrators, router interfaces, etc...).

What else did you want to know?

--
Brad Knowles <br...@his.com> <http://www.his.com/~brad/>
comp.mail.sendmail FAQ Maintainer Emeritus
The comp.mail.sendmail FAQ is now at <http://www.sendmail.org/faq/>
<http://wwwkeys.pgp.net:11371/pks/lookup?op=get&search=0xE38CCEF1>

Jay Lessert

unread,

Oct 23, 1997, 3:00:00 AM10/23/97

to

In article <brad-22109...@brad.his.com>,

Brad Knowles <br...@his.com> wrote:
> Although "regular" ethernet has a theoretical limit of 10Mbps
>throughput, the reality of it is that anything over 27% utilization
>typically starts causing significant numbers of packets to be lost and
>retransmitted, etc.... This is because Ethernet uses a method that
>potentially allows more than one system to be speaking at a time, which
>causes what's called a "collision", leading to both systems needing to
>attempt to retransmit whatever it was they were trying to say and
>hopefully not at the same time (again).
>
> OTOH, Token Ring can pretty much use the full maximum theoretical
>capacity (either 4Mbps or 16Mbps for newer implementations) because it
>guarantees that only one system at a time will be "speaking".

I have enormous respect for Brad's contributions to the sendmail area,
but I'm afraid this is wrong.

The "27%" is even the wrong incorrect number. :-) The old myth spread
by IBM salesman was 37%. But it doesn't really matter, because real,
shared-access ethernets run at 90%+ utilization all over the world
every day.

> The same basic technology used in Token Ring can also be found in
>FDDI, which has a theoretical maximum throughput of 100Mbps (unless you're
>doing something real fancy and probably *very* non-standard). Again, FDDI
>can use virtually all of that theoretical maximum throughput, whereas Fast
>Ethernet (also 100Mbps) can only effectively use about 27% of that
>theoretical maximum (it uses the same technology as "regular" Ethernet,
>just at ten times the speed).

Again, not true, folks like me run 90%+ utilization on shared-access
100BaseT networks every day.

FDDI has does have a small, but measurable throughput advantage for
some types of traffic compared to 100BaseT because FDDI's maximum
packet size is ~4500 bytes, compared to ethernet's ~1500 bytes. This
is only vaguely related to FDDI's token-ring architecture.

> However, very few machines can actually make full use of even Fast
>Ethernet, so that might not be a problem for you.

Any decent modern desktop computer HW, running any decent OS (I'm
including NT here, and even W95, if it is set up by someone skilled in
the art) can saturate a 100BaseT segment. It is even easier for the
same HW to saturate an FDDI segment, again because of the lower
overhead requirements induced by the larger maximum packet size.

Dave Barr

unread,

Oct 23, 1997, 3:00:00 AM10/23/97

to

In article <brad-22109...@brad.his.com>,
Brad Knowles <br...@his.com> wrote:
> Although "regular" ethernet has a theoretical limit of 10Mbps
>throughput, the reality of it is that anything over 27% utilization
>typically starts causing significant numbers of packets to be lost and
>retransmitted, etc....

Cripes, how long has it been since this mythical 27% (or 30%) rule
has been debunked? 9 years? The 30% myth started on Sun
3/50's! Why do people insist on perpetuating such garbage? Things
like the capture effect are now well understood, and is not nearly
the same problem it was when this 30% myth was first introduced.
It's expected nowadays that one can routinely get 70-95% out of an
ethernet network without much trouble.

The fact is also that different protocols react in vastly different ways
when faced with high collisions on an ethernet. NFS, for example, reacts
rather poorly. TCP reacts pretty well.

Must I also mention that most traditional issues are essentially solved
when you go to switched ethernet. (Who _doesn't_ do switched ethernet
these days. It's getting pretty hard to find in most circles.)

http://www.research.digital.com/wrl/publications/abstracts/88.4.html

--Dave
--
http://www.cis.ohio-state.edu/~barr/
ba...@cis.ohio-state.edu

John Doherty

unread,

Oct 23, 1997, 3:00:00 AM10/23/97

to

In article <brad-22109...@brad.his.com>,
br...@his.com (Brad Knowles) wrote:

| Although "regular" ethernet has a theoretical limit of 10Mbps...
| anything over 27% utilization typically starts causing [problems]

Hmmm. 27%. Let's call that about 350 KB/s.

Now let's get some data:

ftp> put test.tif
200 PORT command successful.
150 Binary data connection for test.tif (192.168.1.1,61356).
226 Transfer complete.
local: test.tif remote: test.tif
1338068 bytes sent in 1.4 seconds (9.2e+02 Kbytes/s)
ftp>

That's over a plain old 10 Mb/s thin ethernet. One of the machines is
a sparc-5, and the other is a P133, with a router between them.

So where did this 27% number come from, anyway?

sui...@freenet.edmonton.ab.ca

unread,

Oct 24, 1997, 3:00:00 AM10/24/97

to

Hi Brad:

Thank you very much for your valuable input. It is absolutely wonderful.

Actually, I have problems to handle the network traffic. I have two
servers are for images. One exports images files to another. I am
wondering whether the performance will be slow down for this reason. Where
do I start to look into?

Thank you for your guidance.

Best regards,

----------
From: Brad Knowles[SMTP:br...@his.com]
Sent: Wednesday, October 22, 1997 9:42 PM
Subject: Re: Ethernet Vs. Token Ring

(A copy of this message has also been posted to the following newsgroups:
comp.unix.admin)

In article <62di49$k7e$2...@news.sas.ab.ca>, sui...@freenet.edmonton.ab.ca ()
wrote:

> Can I know the pos and con about Ethernet and Token Ring?

Ethernet is more common in most Unix/PC/TCP-IP environments, wherease
Token Ring is much more likely to be found in IBM Mainframe environments,
but Token Ring has it's advantages for Unix/PC/TCP-IP environments as
well.

<....... Very good stuff deleted.....>

Jason Clifford

unread,

Oct 24, 1997, 3:00:00 AM10/24/97

to

sui...@freenet.edmonton.ab.ca wrote:

> Actually, I have problems to handle the network traffic. I have two
> servers are for images. One exports images files to another. I am
> wondering whether the performance will be slow down for this reason. Where
> do I start to look into?

Have you considered putting the second server on a separate network
consisting of just the 2 servers so that they do not impact upon the
main network and have the best possible performance between the two?

I have been in a similar situation several times and this is often a
good solution unless of course the second server is also being used by
all of the people using the first one.

Jason

Jason Clifford

unread,

Oct 27, 1997, 3:00:00 AM10/27/97

to

ane...@bannerfoundry.com wrote:
>
> In regards to pricing of Token Rings - check out this press release from
> Bay Networks http://www.baynetworks.com/News/Press/nz/nz970613.shtml They
> are talking about a low-cost Token Ring workgroup switch. Are you
> familiar with it?

I am not familiar with this product as such however I would say that Bay
are a well known company and they do produce good products.

However I question the value of such a device in a workgroup environment
which is usually quite a small network where a well designed ethernet is
normally the better choice.

Steve Gibson

unread,

Oct 30, 1997, 3:00:00 AM10/30/97

to

Jay Lessert wrote:
>
> In article <brad-22109...@brad.his.com>,
> Brad Knowles <br...@his.com> wrote:
> > Although "regular" ethernet has a theoretical limit of 10Mbps
> >throughput, the reality of it is that anything over 27% utilization
> >typically starts causing significant numbers of packets to be lost and

> >retransmitted, etc.... This is because Ethernet uses a method that
> >potentially allows more than one system to be speaking at a time, which
> >causes what's called a "collision", leading to both systems needing to
> >attempt to retransmit whatever it was they were trying to say and
> >hopefully not at the same time (again).
> >
> > OTOH, Token Ring can pretty much use the full maximum theoretical
> >capacity (either 4Mbps or 16Mbps for newer implementations) because it
> >guarantees that only one system at a time will be "speaking".
>
> I have enormous respect for Brad's contributions to the sendmail area,
> but I'm afraid this is wrong.
>
> The "27%" is even the wrong incorrect number. :-) The old myth spread
> by IBM salesman was 37%. But it doesn't really matter, because real,
> shared-access ethernets run at 90%+ utilization all over the world
> every day.
>
> > The same basic technology used in Token Ring can also be found in
> >FDDI, which has a theoretical maximum throughput of 100Mbps (unless you're
> >doing something real fancy and probably *very* non-standard). Again, FDDI
> >can use virtually all of that theoretical maximum throughput, whereas Fast
> >Ethernet (also 100Mbps) can only effectively use about 27% of that
> >theoretical maximum (it uses the same technology as "regular" Ethernet,
> >just at ten times the speed).
>
> Again, not true, folks like me run 90%+ utilization on shared-access
> 100BaseT networks every day.
>

No! Both answers *can* be correct, for a typical *shared access* network
of "typical" size, the original post was pretty much correct.
Running CSMA-CD under heavy load might be using close to 100% of the
*bandwidth* but the thruput is *much* less and depends on the network
size and offered traffic.
Do the maths, if a=t/T where t is the end-to-end propagation time and T
is the frame transmission time, then for non-persistent CSMA, the
*MAXIMUM* thruput for a=1 is around 18 %, for a=0.001 the maximum
thruput is around 95%. They are the MAXIMUM values, ie. for a specific
load. loads above this cause the thruput to fall rapidly. Your "a" value
really limits the best thruput you could possibly acheive, once you
average out the load it of course drops!
That's essentially why you can't run ethernet as a MAN, MAXIMUM thruput
would be limited to around 6% for a typical MAN, it's also why the ALOHA
protocol works well for satelite, its performance it independent of "a"
and is's a much simpler protocol.

Of course, if your ether-network is all point-to-point ie. each box is
wired directly to a router, depending on "a" again, you could get quite
high thruput between two machines via the router if you are not sharing
a link and if the link is fairly short you could be up in the 90%'s.
Once you start wanting shared access to a link ie. you have a number of
machines directly connected to a router and the router directly connects
to a server and all the traffic is machine<->server then you are back to
contending for the shared router->server link and the thruput drops off
as above.

So the short answer is it depends on *your* setup, You really need to
know some details about your network and know how these affect the
protocol! The only thing you can state with certainity for *any* CSMA
network, is that one you pass the optimal load, thruput falls rapidly.
Where and how fast depends on your network.

There are a number of fairly standard texts which analyse all this in
detail.
(don't have

hope this helps, SG.
--
Steve Gibson
ADI Limited, CCIS Systems Group
Email: gib...@ccis.adisys.com.au
Snail: 22 Stirling Hwy
Nedlands
Western Australia
Nbrs : (tel) +61 8 9333 8917 (fax) +61 8 9333 8889

Steve Gibson

unread,

Oct 30, 1997, 3:00:00 AM10/30/97

to

Steve Gibson wrote:
>
> Steve Gibson wrote:
> <stuff deleted>

> >
> > There are a number of fairly standard texts which analyse all this in
> > detail.
> > (don't have
>

Hit the send key to fast 8->
should have read;

"(don't have any references at hand, I can dig them out if need be)"

later