| I'm confused.
> The following is the closed-form solution for g=1 (in Maple format):
>
> n-1 n*c
> 1 - (---)
> n
Given the definition of your problem, I would have thought that when g=1
it collapsed to the "birthday problem" - i.e. if there are N days in the
year, and M people in a room, what is the probability that all the people
have distinct birthdays.
The solution to the birthday problem is
N(N-1)(N-2)...(N-M+1)
P(N,M) = ---------------------
M
N
Assuming that you wanted the probability of every caller getting connected,
I would have expected this to be the answer - it has the desired property
that when M (=c*n) > N (=n) the probability is zero. If your formula gives
the probability of some caller NOT being connected, it should then go to 1.0
when the number of callers exceeds the number of called numbers...
So - do you mean something different by "connection"?
Richie
|
| I was imprecise in my statement of the problem. I should have asked for the
average expectation of all the attempted connections, or something like that.
However, with the addition of the factor c, that statement is wrong.
Let me try again, I hope I have better luck.
Given n telephone numbers, each with g lines, and n*g*c customers calling
these numbers, what is the utilization of the n*g total lines ?
Some examples:
For n=2, g=1, c=1, there are two ways that there can be 2 connections and
two ways that there can be 1 connection, for a total of 6 connections in 4
cases. Therefore there is an average of 6/4 connections over 2 lines or
a utilization of 3/4.
For n=2, g=2, c=1, there are two ways to get 2 connections (all four callers
try to connect to the same number), 8 ways to get 3 connections, and 6 ways
to get 4 connections. This is a total of 52 connection in 16 cases. The
utilization in this case is 13/16.
The following table summarizes the above cases and more for n=2, c=1
g
1 2 3 4
N c
u o 1 2 0 0 0
m n 2 2 2 0 0
b n 3 8 2 0
e e 4 6 12 0
r c 5 30 16
t 6 20 56
e 7 112
d 8 70
Total
connected 6 52 324 1768
Total cases 4 16 64 256
The following table (created by enumeration) is for c=1
g
n \ 1 2 3 4 5
----------------------------------------------------------------------------
2 ! 3/4 13/16 27/32 221/256 449/512
3 ! 19/27 569/729 16099/19683 148987/177147
4 ! 175/256 50227/65532 13529521/16777216
|
| This may be what you want:
(sorry for the ugly notation, I didn't have time to prettify it)
let u (user count) = n*g*c for shorthand.
Prob of dialing a given number = 1/n
Probability of exactly m out of the u callers dialing a given number = p(m) =:
uCm * (1/n)^m * (1-1/n)^(u-m) The usual binomial distribution,
xCy = # combinations of x things
taken y at a time.
Expected value of the number of connections on a given number is:
E = sum (m=0 thru m=u) (min(m,g)*p(m)) Since max connections = g
Expected utilization of the lines is E/g.
This formula seems to produce the same results as your table for the cases
in the table. I didn't see a way to express E in closed form.
Richie
|