1
$\begingroup$

I am fairly new to statistics and just recently encountered queueing theory.

I have programmed a simulation for a $M/M/1$ queue in which I specify the inter-arrival times and service times. I input say, an exponential distribution with a mean of $1$ for both the inter-arrival and service time.

I also measure for the effective arrival rate, meaning I run for, say, $1000$ time steps and at each time step a random value is drawn from the exponential distribution. I collect these random values in a list and then compute the effective arrival rate, this being the mean of the list. In theory, this value should converge to the mean, yet, in practice I end up with values not so close to the mean.

My question is, how many random values from an exponential distribution should I draw such that the mean of these converges to the mean of the distribution?

$\endgroup$
0
$\begingroup$

The exponential distribution with mean $1$ also has standard deviation $1$. If you collect statistics for $1000$ time units, you should get a sample mean of about $1$ with a standard error of about $0.032$. When you say your values are not so close to the mean, how far off are they?

By the way, you should be aware that if both the inter-arrival and the service times have mean $1$, the queue is not stable.

$\endgroup$
0
$\begingroup$

Let $\lambda$ be the interarrival rate and $\mu$ the service rate (so the mean interarrival and service times are $\frac1\lambda$ and $\frac1\mu$, respectively). We can model the $M/M/1$ queue as a continuous-time Markov chain $\{X(t) : t\geqslant 0\}$, where $X(t)$ is the number of entities in the system at time $t$, with state space $\mathbb N\cup\{0\}$ and transition rates $$t_{ij} = \begin{cases} \lambda,& j=i+1\\\mu,& j=i-1.\end{cases} $$ To compute the stationary distribution, i.e. $\pi_i := \lim_{t\to\infty}\mathbb P(X(t)=i)$ we have the flow balance equations

$$\lambda\pi_i = \mu\pi_{i+1},\quad i=0,1,2,\ldots $$ This implies that $\pi_{i+1} = \frac\lambda\mu\pi_i$ and by induction, $$\pi_i = \left(\frac\lambda\mu\right)^i\pi_0. $$ Since $$\sum_{i=0}^\infty \pi_i = 1, $$ it follows that $$1=\sum_{i=0}^\infty\left(\frac\lambda\mu\right)^i\pi_0 = \frac1{1-\frac\lambda\mu}\pi_0 $$ and $$\pi_0 = 1-\frac\lambda\mu. $$ Let $\rho = \frac\lambda\mu$. Then the stationary distribution is $$\pi_i = (1-\rho)\rho^i,\quad i=0,1,2,\ldots. $$ Note that this only holds when $\lambda<\mu$. Otherwise, the geometric series above would not converge, and the system is unstable, as @BrianTung pointed out.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.