What is the limit? (was RE: multi-homing fixes)

Wed Aug 29 16:50:01 UTC 2001

On Wed, 29 Aug 2001 12:21:23 EDT, "Martin, Christian" <cmartin at gnilink.net>  said:

> Since I wasn't there during your problems in 1994 and 1996, I ask you, "Did
> you not know that you were approaching a resource constraint?"  In regards

Ever been driving down an unfamiliar road late at night, seen 2 headlights
in the distance, and then been very surprised when you realize that instead
of 2 headlights a normal distance apart a long way off, it's a Jeep with two
very closely-set headlights RIGHT ON TOP OF YOU?

Yes, you know you're approaching, but your estimates of arrival time may be
very poor....

The problem is that we're basically talking about queueing theory, where
the queue is "incoming routing updates".  As anybody who's done a lot of
driving in a major city will tell you, if the traffic is 80% of the road's
capacity, things are good, at 90% they're good, at 98% they're tolerable, and
at 101% you're screwed.

Traffic engineers (the highway kind) have long known about the equivalent
of "ringing" on major highways - if there's a temporary congestion, there's
a backlog.  When the congestion is cleared, you end up with a traffic jam
moving backwards down the road, as cars at the front come back up to speed,
but cars arriving at the back have to slow down.  Such backwardly mobile
jams have been observed to go around the Washington DC beltway 2 to 3 times
before finally dissipating.

The difference is that the highway traffic engineers have close to a century of
experience, and have gotten very familiar with the behavior patterns of
congestion (take 30 major cities in the US, and in 1 year you have 10,000
examples of meltdown-level congestion to base your numbers on, assuming that
each city only has 1 traffic jam per day).  We've been at it in a serious way
only a decade or two - and we've intentionally avoided meltdowns (notice that
Sean Doran only cited 2 examples for a decade).  This means we're on a lot
shakier ground when trying to make predictions - we *dont* have tens of
thousands of data points....

-- 
				Valdis Kletnieks
				Operating Systems Analyst
				Virginia Tech

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 211 bytes
Desc: not available
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20010829/1074e6a4/attachment.sig>