# calculating how much to oversell given an acceptable risk (statistics)

I have a shared resource with a finite capacity (let's say 100), and I have usage data (2 hours average of samples taken each 20 seconds). I accept a risk of 10% per year to reach the capacity.

Knowing how many users/samples I have, and assuming all users follow a similar distribution of use (which usually is not uniform (ie, in some periods the usage is higher than the average for almost all users)) and that they have a maximum allowed usage (for instance, 5), is it possible to calculate with how many users I can share this resource?

I though this was a very common problem, but I couldn't find many solutions on my researches.

What I'm currently doing is calculating the mean/variance of the usage and the p-value of the resource after I put a new user in. If it's higher than 0.10/year I don't do it. I see some caveats like it's not considering the fact that the 2 hours averages are from samples taken each 20 seconds; and it also doesn't assume that a user can never use more than his allowed usage.

Any hints on what I can do and/or where to find more information?

thanks

• Can you describe the setup more precisely? It is not at all clear to me what the usage model is. I don't even understand the phrase "10% per year" :(. – fedja Feb 14 '13 at 5:46