[IP] Quick ?for list

[David Farber <dave@farber.net>]

________________________________________
From: Dave Burstein [daveb@dslprime.com]
Sent: Friday, February 15, 2008 4:15 PM
To: Steven M. Bellovin; David Farber
Subject: Quick ?for list

Dave

Steve's asking the right question, but with a confusing context. If you pick it up (and anything from Steve would normally be worth passing on), I suggest you include the short bit below for clarity.  I've a longer answer coming to both of you shortly.

Steve asks whether looking at all the carrier costs makes a difference, which he notes is a very small part of the ISP's total cost.  For this purpose, open access/net neutrality/traffic management, the bandwidth costs is essentially the only cost that changes whether you use 100 kilobytes or 100 gigabytes. The difference in cost between a neutral and effectively non-neutral network is probably closer to ten cents than to a dollar. The question on the table is whether this is "reasonable network management" or not, and whether imposing a burden on Comcast of (ten cents to one dollar per/subscriber, based on assumptions) is "reasonable" policy. Opinion can differ about that (I believe so), but I don't want to review the debate here.  Most useful datapoint on this is the carrier cost of a gigabyte was about ten cents and has probably gone down. Ask offlist if you want sources or more details, and especially if you have data that suggests I have a fact wrong. Dave Burstein

> Over that same period, the costs of delivering that bandwidth have
> gone down at a Moore's Law pace of 25-40% (switches, routers, etc.)
> The result is that the carrier's cost of bandwidth has been flat to
> down for five years. The total bandwidth cost is typically
> $1/month/customer. Multiple sources
>

This is an interesting question: what is the cost of bandwidth?  My
impression is that most of the cost of running an ISP is personnel,
rather than cables or fibers, routers, etc.  Clearly, this will be
different for something like Verizon's FIOS or AT&T's Lightspeed, where
there's a hefty capital cost to amortize over some number of years.
Most of the hard part about deploying IP over HFC (hybrid fiber-coax)
was done years ago; accommodating higher bandwidth is largely a matter
of deploying new fiber nodes and splitting the trees, which is a cost
that would be driven down by Moore's Law and economies of scale.


                 --Steve Bellovin, http://www.cs.columbia.edu/~smb

-------------------------------------------