I just realized the reason why Dave Clark's in- or out-of-profile
packet marking and selective dropping model is the Right Thing -- I
mean I believed it before, but I wasn't sure it wasn't just because I
hated the alternative of paying for all my bits.
Profiling reflects the fundamental underlying economics of data
transmission: the individual fixed-capacity connection. In the end,
the phone line / satellite connection / cellular voice channel all
share a basic step-function pricing rule: below capacity, the marginal
cost of additional bits at that instant is zero, and above it's
infinite. Aggregators -- firewalls, ISPs, backbone providers -- have
more gentle curves because they can shape overall traffic, add
channels, etc. Most importantly, their 'additional costs' are actually
incurred per-packet: routing equipment, primarily. That's the
per-usage argument. But in the end, any channel I can think of off the
top of my head is really fixed-capacity and exhibits this pricing.
Nevertheless, I was initially skeptical of in- or out- 'pricing' in
Clark's work. Information is always valued in shades of gray, right?
But, after thinking about it recursively, it eventually ends up being
applied at the router port actually connected the backbone
channel. There, on the edge of the DS3, lo and behold, the same
'pricing' algorithm is at work: either the next packet can be shoved
on the line for free, held for a little while (incurring routing
costs), or thrown away.
Packet-coloring merely extends this rule backwards, back to the end user
who's actually contracting for a 'softer' version of the same service:
paying for a 'virtual' channel of X bits per second at Y latency with
Z reliability. The end user has to color their packets according to
whether they're within the contracted limit. If there's excess b/w
(because, after all, aggregators take advantage of the law of large
numbers), the excess packets go through at their true economic price:
zero. And if not, the end-user also ends up paying the true economic
price: it gets dropped.
Rohit Khare
PS. I make no claim this is original thinking; it's probably plainly
obvious to the most casual observer. It helped me, that's all.
Rohit Khare -- World Wide Web Consortium -- Technical Staff
w: 617/253-5884 -- f: 617/258-5999 -- h: 617/491-5030
NE43-344, MIT LCS, 545 Tech Square, Cambridge, MA 02139
=================================================
Date: Tue, 25 Feb 1997 17:07:45 -0500 (EST)
From: Rohit Khare <khare@w3.org>
To: FoRK@xent.w3.org
Subject: Re: Freeway or Feeway?
Cc: andyo@ora.com
http://webreview.com/97/01/31/trends/index2.html
Andy Oram, an O'Reilly book editor, also did a report on that MIT
net-economics panel with Varian, Clark, and Metcalfe I minuted at:
http://xent.w3.org/FoRK-archive/winter96/0232.html
Rohit Khare
Rohit Khare -- World Wide Web Consortium -- Technical Staff
w: 617/253-5884 -- f: 617/258-5999 -- h: 617/491-5030
NE43-344, MIT LCS, 545 Tech Square, Cambridge, MA 02139