[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Another Point Of View (Round 2)
At 01:47 PM 11/29/96 -0500, Bill Frezza wrote:
>Voice traffic follows a classic exponential, or Poission, distribution. It
>is the basis for the Erlang tables used by phone companies to provision
>trunk groups. Data calls *Do Not* follow a Poisson distribution. See
>"Impacts of Internet Traffic on LEC Networks and Switching Sysetms", Amir
>Atai and James Gordon. (Note that nomenclature for Probability Density
>Functions is not uniform. Perhaps you learned about Power Law PDFs under a
>different name. Or maybe you were absent that day :)
Voice calls don't exactly follow the classical Poisson exponential
distribution; they're just close enough that the models (Molina and Erlang
B) work pretty well. Data calls don't generalize all that well. The
distribution varies depending on the type of call. I've noted bimodal and
trimodal effects. Averages are somewhat longer than voice, but I don't
really think that it matters much -- IF you discount "nailed" calls, which
*are* a problem, then it's close enough to Poissonian to work quite well.
>> BF> Ignoring the fundamentals of traffic engineering, CPT argues
>> BF> that a 20-minute local data call costs the phone company the
>> BF> same as a 20-minute teenage yak-fest,
>>This isn't true?
>Of course a 20-minute call is a 20-minute call regardless of whether it's
>voice or data. But that's not the point. The phone network is not
>engineered to handle 20-minute average holding time calls. In addition,
>it's not the *average* call that matters, it's the total call-minute load
>presented by the entire distribution function.
AHA! This last paragraph is oxymoronic. First Frezza argues (in his
column) that 20 minutes isn't 20 minutes. Then he says (to Mike Bilow) that
20 minutes *is* 20 minutes ("it's not the *average* that matters"), but that
the total call-minute load matters.
Well I agree with his last statement. Total minutes-per-month is the
measurement that matters.
Now the phoney-baloney Bell Atlantic argument is that ISP incoming lines
have heavier-than-average usage. Of course they do -- they're large hunt
groups! You don't put in 500 lines for 200 peak calls. I, for one, use
ordinary Poisson tables (well, plus some off-the-top extrapolations) in my
day-to-day dial-in data traffic engineering. As a first-order approximation
for multiple- PRI-size groups, they allow busy hour loading to be around 92%
before blocking goes gaga. Now matter how much you jigger the formulas,
you're not going to make more than a few percent difference from that number.
Frezza also mis-states what I understand the CPT position on R-ISDN to be.
They are aware of the nailed-line problem, and accept threshold rates
("caps" on flat rates) in the 200 hour/month range. These allow just about
any "normal" usage while discouraging nailed use. They also support ISDN
rates that cover telco's costs, with fair profit margins. Discounting
nailed users, average Residential ISDN usage is around 20-30 hours/month.
Bell Atlantic's own numbers (the "expurgated" ones) show an average
*traffic-sensitive local usage* cost/minute in the 0.3 cents range, say
15-20 cents/hour (per B channel). So 30 hours at 20 cents is six bucks. I
say let'm have ten for a 200-hour (sum of two B channels) usage plan, more
for "extended mileage" plans. And for nailed users, well, let'm pay their
own way, which will be a *lot* more. (You do the arithmetic.) See, I'm a
sport, willing to pony up fair costs. If that's what telcos were asking
for, then I don't think CPT would be arguing about it so much either.
Fact is, the Internet has been a fast-growing new appplication for the phone
network that the phone companies didn't anticipate. Given their glacial
planning cycles, they're at a loss, so they're trying to shuck the whole
thing. Sorry, but I think they can do better.
Fred R. Goldstein k1io firstname.lastname@example.org +1 617 873 3850
Opinions are mine alone. Sharing requires permission.