Re: Station wagon full of bits


Phil R. Karn (karn@flash.bellcore.com)
Wed, 25 Mar 87 22:16:29 est


You really should update this old idea to use more up-to-date technologies.
To wit:

What is the bandwidth of a B-747 (DC-10, L-1011, A-300) loaded with Compact
Discs or CD Roms?

(To be fair, we should compare this to a fiber link covering the same path.)

With fiber, the economies of scale in transmission are enormous. However,
the ARPANET has been around a long time. Its internal routing algorithms
were built to squeeze the most out of the slow and expensive leased lines
available in 1969. Its internal protocols and congestion control techniques
were designed to squeeze the most out of the expensive, memory-poor
computers available in 1969. It was an excellent computer network -- for
1969. Unfortunately, much of this simply can't scale in its present form to
where it can effectively utilize the new economies of scale in transmission.

Making decisions about the future of networking by comparing the "operating
costs" of ARPANET and MILNET to GTE Telenet is dangerous. For one thing,
Telenet was also built on 1970's technology and protocols. Also, its users
pay *real money* depending on how much traffic they send. Unlike the
ARPANET/MILNET, Telenet's primary use is as a nationwide "remote terminal
concentrator", with the vast majority of users dialing into X.3/28/29 PADs
with dumb terminals and connecting to timesharing hosts. Their only
alternative is to dial a direct long distance call using AT&T or MCI or
whatever, and Telenet is well aware of this; they price their service
accordingly. Anyone who tries to use it for computer-to-computer
internetworking (as we do through CSNET/X.25NET) finds out VERY quickly just
how expensive this can be. The circuit-switched mentality is deeply
ingrained in Telenet's internal design; at least with the DDN, X.25 is kept
on the edges, making it at least theoretically possible to escape its
braindamage.

In deciding where to spend future bucks, I think it would make a lot more
sense to look at newer technologies. Because the transmission guys have
leapfrogged so far over the switching guys, a radical change in mindset is
in order. It simply doesn't pay anymore to worry about efficient PSN buffer
utilization, 100% delivery reliability, packet header overhead, finding the
most optimal routes, or load splitting if doing so takes so much CPU time
that you can't drive fast links at full speed. At present, only link-level
bridges like the Vitalink Translan III seem able to switch packets quickly
enough to drive a T-1 link to 100% utilization on small packets; if somebody
can do this with an X.25 switch or IP gateway I would like to know about it.

Phil



This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:37:45 GMT