More BW, Less Taxes

Sean M. Doran smd at
Tue May 29 16:48:08 UTC 2001

| It might make more sense to ask "Why is bandwidth so cheap (and getting 
| cheaper so fast) in the US?" The simple answer is: Moore's law, competition 
| (leading to a "fiber glut") and economies of scale.

ignoring detours into the land of atm/fr/x25/etc-under-IP:

1. ripping out t1 channel banks and using t1s directly for ip
2. ripping out m13 muxes and using t3s directly for ip
3. ripping out all sorts of weird t3 gear and using sonet (oc3->oc48)
   direclty for ip
4. ripping out sonet gear and using sonet-framing between unsynchronized
   pairs of routers over wdm systems built with sonet in mind

this has made things cheaper and easier to manage over time,
so the cost of supplying each chunk of 64kbps has gotten cheaper
over time.  result: fewer $ per mbps of p2p bandwidth for router

note that long haul individual DS0s are really not that much cheaper
now than they were a few years ago.

on the ip side, in a vendor-specific timeline ordering:

1. better forwarding algorithms ('fast switching')
2. multi-chip specialized forwarding engines
3. aggregation 
4. "subnet of small routers" linecard+backplane model of distributed
5. specialized silicon getting cheaper (this is your "Moore's law" comment)

on the "fibre glut" front there are parallel evolutions:

I.   fibre-based networks constructed for inhouse bell-head-apps
II.  "patchy" builds mitigated by cross-selling/leasing arrangements
III. more companies lighting up "patchy" regional networks
IV.  pipeline and other right-of-way companies turning on sizable "patchy" 
V.   companies which "exist to dig" - developments in fast-digging, etc.

many of II-V sell/sold towards the original I companies, as well
as startup plays who do bell-head or even Internet focused optical subnets.

and perhaps even more importantly, in parallel:

1. WDM
3. EDFA, Raman etc.
4. Ultra-long-haul optics with tight filters

which competes with the "more fibre" approach as more bandwidth is required.

| If you're across an ocean from the US, you have to factor in the cost of 
| running underwater cable.

Having done this, I can say that in general the tricky part is mostly on land,
and in the selection of not-so-bell-head-focused equipment & contractors.
Thanks to points 3 and 4 in the last set (optical stuff) above, the
wet pieces in future fibre builds aren't really much more than strings of 
glass that adapt reasonably well to bending and sharkbite.  Digging is
probably the expensive piece in future, followed by paying off the
the people who can charge you for the use of the land you're dropping stuff into.


More information about the NANOG mailing list