[geeks] FDDI

[Dave McGuire]

On Tuesday, November 5, 2002, at 08:29 PM, Lionel Peterson wrote:
>>    Metcalf was WRONG.
>
> Seems kinda harsh...

   I'm a harsh person.  For about three years back in the late 80s/early 
90s, I worked for Gerald Clancy at a tiny three-man (well we peaked at 
five I guess) startup company in Trenton NJ called Princeton Desktop 
Systems, Inc.  He was the guy who chaired IEEE committee 802.3.  He 
doesn't have anything good to say about ethernet either, or Bob Metcalf 
for that matter.

> In the context of being 25 year old tech, it seems OK to me... But then
> again, I think you need to remember the nature of networking &
> workstations "back in the day"...

   Yes, it was great twenty years ago.  Unfortunately, that time has 
passed.

> Not a defense (becuse I speak from ignorance), but a plea for an
> elaboration, if you please!

   Well...I am not a statistician, so I don't know the proper 
terminology here...but consider CSMA/CD.  A station wants to 
transmit...It looks for carrier, if it sees none, it transmits.  If two 
stations decide to transmit at the same time, the packets collide, 
which is detected as noise on the wire.  All stations see their own 
packets on the wire, so if they notice their own packet got trashed, 
they assume it's due to a collision, and wait a random period of time 
and retransmit.

   Well if you think of this over time, the more stations you have and 
the more packets they try to transmit, you not only have more 
collisions, but also a higher the *likelihood* of 
collisions...resulting in yet more collisions.

   Combine this with the ridiculously small packet size of 
Ethernet..1500 bytes.  Smaller packets mean more packets, and more 
packets mean more potential collisions.

   This is why an unswitched 100Mbit ethernet rarely gets more than 
55-60Mbps of usable bandwidth.

   In order to get around this problem, we use ethernet switches to 
eliminate (or reduce the size of) the collision domains...greatly 
improving performance in some situations.  This works fine, you might 
say, so what's the problem?  Well, what we've done here is changed one 
of the very basic design principles of ethernet in order to make it 
perform acceptably.  That boils down to poor engineering in my book...a 
design that hasn't stood the test of time, and needs to be modified at 
a VERY fundamental level (yes this is important) in order to remain 
applicable.  And let's not forget that 1500 byte packet size.

   Then, people go and scale it to 100Mbps, then 1000Mbps...and at least 
1.5yrs ago, I ran across the draft spec for 10,000baseFX.  Thankfully, 
the 1Gb ethernet standard allows for larger packets than ethernet's 
original wimpy 1500 bytes...but not everyone seems to have bothered to 
implement that feature.

   Ethernet is like IDE...It's popular because it's cheap and easy to 
build, not because it's good.  There are newer, better technologies for 
smart people to use...like FDDI and HIPPI.  And in the case of FDDI, 
it's not "suit fashionable" so it's cheap for us people-in-the-know to 
get our hands on.  Those Cisco WS-C1400 FDDI concentrators, for 
example, were $20,000/ea back in 1996 when they were new...we bought a 
bunch of them at Digex.  Now we're getting them for $20! 8-)

            -Dave

--
Dave McGuire                  "Mmmm, big."
St. Petersburg, FL                -Den