IP: 100% per year, etc

[Dave Farber <farber@cis.upenn.edu>]

>Date: Tue, 28 Nov 2000 06:26:22 -0500 (EST)
>From: mo@UU.NET (Mike O'Dell)
>To: farber@cis.upenn.edu
>Subject: 100% per year, etc



>Dave,
>
>I see people still don't really understand the difference
>between offered load (measured as gigabits injected into
>the edge of the network) and network capacity (measured in
>gigabit-route-miles of trunking).
>
>This is indeed somewhat subtle and possibly counter-intuitive.
>
>For offered load to double every year, network capacity
>must double every 4 months or so, at least in our network
>(UUNET).  It is slowing down a some, but that's still pretty fast.
>
>This is actually a pretty simple result from graph theory,
>once one gets the picture right (as are most results
>from graph theory - grin).
>
>I admit it took me a while to get the picture right,
>especially about how to explain what's going on.
>
>Consider an network of two nodes connected by one mile
>of fiber.
>
>In this network, it's pretty clear that the number of
>gigabits of offered load should be equal to the number of
>gigabit-miles of trunking between the two nodes.
>
>Note that it's only the NUMBERS that are equal - the UNITS
>of the numbers are different.  One is gigabits/sec, the
>other is (gigabits/sec)*miles.  This means they can never
>be *equal*, but the two numbers can grow together - maybe
>even related by a constant of 1 mile, as in this case.
>
>Now consider a real network.
>
>The network is much more complex than two nodes with one
>link, and each of those nodes is generating traffic which
>can go to any arbitrary destination node on a
>packet-by-packet basis.
>
>In North America alone, there are "35 NFL Cities" which
>account for a significant fraction of the population
>(people or computers, take your pick), but there are also
>many more computers located other places who also wish to
>have high-performance service.  This means the trunking
>must go a lot of places and be richly connected because...
>
>Every computer expects to reach every other computer with
>nearly equal quality.  While Warhole's Theorem is still
>relevant, on the Internet you are famous for 15 milliseconds,
>so who is famous changes very quickly.
>
>The resulting traffic slosh can be very large and the
>network trunking capacity (gigabit-route-miles) better
>be in place to handle it or Bad Things(TM) happen.
>
>The result is that for UUNET's network (I can't speak for
>others) to handle the 100% increase in gigabit/sec offered
>load over 12 months, the gigabits/sec-route-miles capacity
>of the network must increase 100% about every 4 months.
>Again, note the difference in the units of those two
>numbers.
>
>The planning problem for telephony networks is rather different
>because the dynamics of telephone calls are so much slower
>and the data rate required is perfectly predictable.
>
>I dare say that if 800 numbers were routinely "SlashDotted",
>the outlook of voice network planners would be rather different,
>and that's ignoring the huge difference in bit-mass moved
>in the two cases.
>
>The deep intuition about network growth dynamics developed
>over the years with voice networks simply does not yeild
>workable results when applied to very large data networks
>which exhibit huge dynamic ranges of traffic slosh and the
>astounding doubling of offered load every year.  (and this
>is still the case even given how few people currently enjoy
>"broadband" access)
>
>So the statements about growth of offered load and growth of network
>capacity (no matter how much "ballyhooed") are not inconsistent.
>They are taking about two different but highly interrelated things.
>
>         cheers,
>         -mo
>
>========================
>Michael D. O'Dell
>Senior Vice President, Chief Scientist
>UUNET - the Worldcom Internet Company
>Room E1-3-117
>22001 Loudon County Parkway
>Ashburn, VA 20147
>Voice: +1-703-886-5890
>Fax:   +1-703-886-5806
>Email: mo@uu.net



For archives see: http://www.interesting-people.org/