Real ops talking to future ops

Tue Aug 24 15:11:21 UTC 2010

On 8/23/2010 6:39 PM, John Kristoff wrote:
> A few classes ago I had a student tell me they had an instructor spend
> two full classes (out of 10) on Token Ring.

There's a serious need to cover such a construct, but also to introduce it in 
the context of modern systems:

      Probably none of what is sold today as ethernet is actually the original 
ethernet protocol or even close to it.

      What is sold today is a the ethernet *interface* and some other protocol 
under it.

This difference between the interface and the infrastructure under it that 
provides service to it is a fundamental construct that is often missed. 
Standardized interfaces let technology adapt underneath it.

So, for example, IBM published the API for netbios, without publishing the 
protocol.  That let some of us build alternative protocols that satisfied the 
API but ran over TCP.  (See RFC 1001, 1002 for the standardized version.)

Much of what is sold today as ethernet has a protocol under it that is 
contention-free.  The different Token Ring schemes provide that in a distributed 
manner.[1]

d/

[1] Though my own focus was on email, my CS prof was Dave Farber, so I had to 
absorb more about TR than I would have wanted.  One of the interesting metricts 
for TR is delay-time per node.  The Irvine Ring introduced one bit-time delay. 
Scaled great. The IBM TR introduced one full packet-time.  Didn't scale well.

-- 

   Dave Crocker
   Brandenburg InternetWorking
   bbiw.net