Odds of rolling $K$ or fewer of a particular die face with $N$ dice.

63 Views Asked by At

I have a real world problem that boils down to this:

  • I have $S$ sensors that each send a message once every two minutes to the server. Each sensor sends its data at a time uniformly at random within the two minute period, and is independent of all other sensors.
  • The server can only receive $R$ messages per second (if more are sent errors result)

How large must $R$ be to ensure that no more than $R$ messages are received within one second 99.995% of the time?

Equivalently, if I roll $S$ dice, each with 120 different faces, how large must $R$ be so that 99.995% of the time no more than $R$ of any given face comes up?

1

There are 1 best solutions below

1
On BEST ANSWER

Your description suggests the use of the Poisson distribution:

$$P(k \text{ events in interval}) = e^{-\lambda}\frac{\lambda^k}{k!}$$

In this case, the interval is $1$ second, and $\lambda=\frac{S}{120}$.

You want to find the smallest $R$ such that $\sum_{k=0}^RP(k)>0.99995$, that is, with an exception on average once every $5\tfrac12$ hours.

Now, the cumulative distribution function is $P(k\le R) = P(k<R+1)=Q(R+1,\lambda)$ where $Q$ is the regularized gamma function.

Suppose there are $S=80$ sensors. Then by WolframAlpha:

  • GammaRegularized[6,80/120] $\approx0.9999309$
  • GammaRegularized[7,80/120] $\approx0.9999935$

So $R=6$, which is to say that fewer than $7$ messages will be received within one second $99.99935\%$ of the time, with an exception once every $43$ hours or so.