I've followed with interest the discussion about realtime billing,
anyway, even if it could be a fascinating subject as a developer, I've
always felt that from a project management point of view the problem is
simply non-existent, because the money lost with wide-grained control is
unimportant against the higher system complexity and time spent in
research to achieve fine-grained control.
I've tried to put a couple of formulae on paper to be less qualitative
and more quantitative, be warned that :
1) I don't came from a TLC background, maybe there is something about
call flows that I'm missing
2) I'm not a "businessman", maybe there's something about money flows
that I'm missing
There are 3 ways to control customer's credit, different implementations
have been proposed, but basically :
- "pre dial control", the simplest :
if credit < 0 do the Dial
else redirect to the Playback(sorry, money makes the world turn)
on hangup update credit
- "set maxtime per call", still easy to implement, a bit more heavy on
the cpu, since you have to do some math before the call
if credit > 0 find the price for requested number and set maxtime,
do the call
else Playback(go away, loser)
on hangup update credit
- "set maxtime per call flow", complex to implement, heavier on the cpu
since it needs some kind of polling, heavier on db if you don't cache
credit data
set maxtime per call
then during the call :
every n minutes do :
for each call in the flow (one customer, one flow):
credit = credit - (n * call price per minute)
update maxtime
on hangup update credit
Since this is becoming long I'll go quickly to the point :
the max money you can loose with "pre dial control", which is
obviously the more wide-grained, is
2 * MaxConcurrentCalls * MaxCallLength * medium_price_per_minute
I dont' take in account MaxPricePerMinute since high international
prices are way too higher and less frequent than the average ones.
Given MCL = 1 hour (if you don't set this value chances are your
provider is setting it) and MCC = 20 (that's my biggest customer limit,
and the average is 15/18) we loose around 100$, supposing high prices (
I'm actually loosing 72 euro ). This seems to support my feeling, just
update the credit in a cronjob every 30 minutes or so and you should
sleep safe.
for the average case :
2 * average_calls_per_update_interval * average_call_length *
medium_price_per_minute
now we can infer average_calls_per_update_interval and
average_call_length with select in the db, but I suppose there is some
function like
average_calls_per_update_interval = f(update_interval,
hour-of-the-day)
and my new feeling is that inserting different update_intervals in
different hours-of-the-day in your crontab, so that the above formula
stays below a value choosen with the investors, you can minimize the
extra cpu usage while keeping the system simple and achieve results
equivalent to the ones of more complex solutions.
(as they say in books, I leave the Max/avg formulae for the other cases
as an exercise for the reader)
_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --
Asterisk-Users mailing list
To UNSUBSCRIBE or update options visit:
http://lists.digium.com/mailman/listinfo/asterisk-users
