[e2e] congestion pricing in action...
rbriscoe at jungle.bt.co.uk
Sat Apr 13 05:02:08 PDT 2002
At 18:04 25/03/02 -0500, Bill Sommerfeld wrote:
>Seen in another context, and forwarded largely for amusement..
The serious point here is that they aren't using congestion pricing down to the granularity of the service (i.e. packet granularity), so they leave themselves open to pathological behaviour like this. A classic example of incentive incompatibilty. ECN pricing doesn't suffer from this flaw. I made this granularity point in the paper below:
Title: "Lightweight, End to End, Usage-based Charging for Packet Networks"
Author: Bob Briscoe and Mike Rizzo and Jérôme Tassel and Konstantinos Damianakis
In: "Proc. IEEE Openarch 2000", pp. 77--87
Date: March, 2000
>[The provider in question measures usage with 5 minute granularity, and
>charges extra if the 95th percentile bandwidth exceeds a threshold..]
>Subject: bandwidth hogging schedule
>Date: 23 Mar 2002 01:39:09 -0500
>Due to the way <deleted> does billing, once we burst over our 1/4
>meg bandwidth limit in any five minute period, there is no additional
>cost to really leaving that limit in the dust :-)
>In an average 30 day month, we can burst over our limit for 432 five
>minute slices without any extra charges. This averages about 14
>slices per day.
>So, I am declaring six of those slices each day, totalling 30 minutes,
>to be the collective bursting window. If you have any cron-executed
>backups, large transfers, or other bandwidth-needy tasks to run,
>please set them to run between 4:31am and 4:59am. The extra minute
>missing on either end is so those of you with poorly synced clocks
>still land in the window.
>This leaves plenty of additional bursting time for when random use
>causes us to exceed the 1/4 mbps threshhold.
Bob Briscoe http://www.btexact.com/people/briscorj/
More information about the end2end-interest