Template Tools
For the record: Bandwidth Costs
Thursday, 24 June 2010 18:20
There's no news here because I have already reported all of this, but in the spirit of disclosure I like to report important comments I've made to FCC and other officials.
    A senior lobbyist presented to the FCC data I considered highly unlikely, so I sent some conclusions over to the FCC. The lobbyist - who I won't embarrass by naming here - filed a claim of  "annual traffic growth rates of between 100% and 114% through 2014." In addition, the projected the cost of handling that bandwidth "decreases at 10% per annum over the 2009 to 2014." Cisco and nearly everyone I know predict growth rates of 20-40%, with Cisco's respected VNI predicting U.S. growth will fall to less than 20% in 2014. (Hard to tell.) Senior officials of both Cisco and Alcatel tell me they expect declines in the cost of equipment needed for bandwidth (switches, routers, DWDM, etc.) will "continue on the historical trend or possibly decline slightly faster." (25-45%.) Combine the two errors and the projection is off by a factor of about 10 and distorting any discussion of "managed services."

    When I saw the filing with the data I questioned, I sent this note over to the FCC:

    Bandwidth is not free but it's cheap enough to almost never be an issue at large DSL, Cable or fiber. I think that's mostly true on wireless with careful planning, but I'm not a wireless expert.
Quick numbers:
Current provision per customer, no congestion noticed: 100K (per Alcatel and about on target.)
Current cost per megabit, large carrier, all in. ~ $10. (Transit is down to $2 to $5, plus upgrades needed to switches routers and occasionally DWDM from the peering point to the local exchange.) (Alcatel used a $25/megabit figure when Cogent, for example, is selling at $4/megabit. NTT told me on the record that some recent very big contracts came in at $1.50-$2.)

Net cost of bandwidth per customer, today: $1/per month.
This figure is the industry standard and has been roughly stable for 5-8 years. Bandwidth demand has been growing at 30% (AT&T figure) to 40% (Comcast out of date figure) for each user. Switches/routers/transit have been getting cheaper at 25-45% per year driven by Moore's Law.

Likely future cost: Almost certainly similar.


Both Cisco and Alcatel are confident switch prices will remain "on the historical trend or better." They are racing to get their 100G gear to market, with Ciena/Nortel already shipping.

Cisco's future bandwidth projections, the most carefully done, are that U.S. growth in bandwidth per user will go below 20% by 2014. That's totally unproven.

Result: Costs go down 25-45%, bandwidth demand goes up possibly slower.
Cost per month per customer should be flat to down as far out as we can look.

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

The only interesting exception is increasing AT&T and other telco capacity beyond what the copper line in place can handle i.e. replacing DSL with fiber to the home. Digging fiber is expensive. DSL speed is going up 10-30% with management of noise and will probably double in new builds in 3-5 years (vectoring to cancel noise) but there's a limit. But since the chances of AT&T, Century, Qwest or other telco doing large fiber home builds in the next five years is nil, I wouldn't put that into a policy discussion.

Implied in that is cable's ability to increase total speeds to maintain the 50-100 meg for each user. John Chapman of Cisco at the Cable Show presented a paper on how to get to 1 gigabit, shared, at reasonable cost. The cable guys can probably keep the 50 meg speeds solid through 2020 while reducing capex, as they are.

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

I can provide solid backup for all of this if any is in doubt, from the top technical people at the carriers.