Expected value depending on $k$ identically independent exponential distributions

47 Views Asked by At

Suppose a computer has $s$ processors with identical independent exponential processing times with rate $\mu$. Instructions are processed on a first-come first-serve basis as soon as a processor becomes free. Instructions arrive with independent exponentially distributed inter-arrival times with rate $\lambda$.

Suppose that an instruction finds itself first in line with all processors busy. Denote $W$ the waiting time of the instruction and $T_P$ its processing time once accepted by a processor. Let $T = W + T_P$ be the total time elapsed between the arrival of the instruction and the completion of its processing.

I need to find the distribution of $W$ and calculate $\mathbb{E}[T]$.

My Thoughts

I've found several documents that state that the expected value of the waiting time of a customer in the queue is $$W_q = \left[ \frac{1}{(s-1)!}\left(\frac{\lambda}{\mu}\right)^s\frac{\lambda \mu}{(s\mu-\lambda)^2}\right]p_0,$$ where $p_0$ is the probability of $0$ customers in the queueing system. Now how can this expected value help me to find the distribution of $W$?

1

There are 1 best solutions below

0
On BEST ANSWER

This is much easier since you know the exact state on arrival of this particular instruction. Since it's first in line and all the processors are full, the waiting time $W$ is the time until the first of the $s$ instructions being processed is done. Since all is exponential/memoryless, the wait time till the first finishes is exponential with rate $\mu s.$ (This is from the superposition principle for Poisson processes, or more simply from the fact that the minimum of $s$ independent exponentials with rate $\mu$ is an exponential with rate $\mu s$.)

Thus we have $E(T) = \frac{1}{\mu s}+ \frac{1}{\mu}$