Re: Very fast IP, no ATM...

[Petri Helenius <]

Jon Crowcroft writes:
 > 
 > weel, for toll quality speech, we can get down to around 
 > 16kbps per user, i.e. 4 per 64kbps, i.e. nearly 100 per 2Mbps, or
 > 30,000 per 622Mbps...
 > 
 > now assume that out of the worlds 600M phones, 10% are active

This might be realistic but is actually on the higher side.
 > 
 > and assume that most calls (say 90%) are local
 > then we need to tolerate a total of 6M calls....(this is WAY higher
 > than they actually do...)
 >
Exactly. The average active international call count for 3 million
finnish telephones is less than a thousand. That would make the
international call percentage 0.033% of active call minutes which
sounds realistic.

 > assume that we build a path disjoint backbone, so continental locality
 > is also kept reasoanbly well (and timezones help anyhow) - we need to
 > support 600,000 calls  on an Internet trunk...
 >
And with the above more realistic figure, this number comes down to
20000. Which in order sounds still high knowing the infrastructure in
existence to carry the calls. 

 > so we are out by 20 times right now.....
 >
Take a short look how many bits per second is actually in use between
the continents and then figure out where the phone calls must fit :-)

 > ok, so we put in a 16 fold replicated backbone global grid...then we're there
 >  
 > but there are 800k web browsers in the UK after our interntational
 > links anyhow, so we need to upgrade anyhow...
 > 
 > so they'd all suddenly get instantaeous reponse...
 > 
 > oh, the cost?
 > 
 > tricky one:-)
 > 
When we get more privately funded intercontinental cables like FLAG,
the availability of bandwidth for a reasonable price will be
here. Until then, we've just wait for the internet-aware cable
investors to pop up.

Note that it takes half a decade from a plan to a cable. 

Pete