Invariant Distribution of Two Dependent $\cdot/M/\infty$ Queues Running in Parallel

71 Views Asked by At

This is in preparation for an exam I have coming up.

We have two $\cdot / M / \infty$ queues with external arrivals occurring according to a Poisson Process of rate $\lambda$. Service occurs with rate $\mu$.

We are told that if an arrival comes to find $n_1$ customers in Queue 1, and $n_2$ customers in Queue 2, it will join Queue $j$ with probability $\frac{n_j + 1}{n_1 + n_2 + 2}$.

I am asked to show the Markov Process $N(t) = (N_1(t), N_2(t))$ has invariant distribution $$\pi(n_1,n_2) = c\frac{\rho^{n_1 + n_2}}{(n_1 + n_2 + 1)!}$$ where $c > 0$ is some suitable constant and $\rho = \frac{\lambda}{\mu}$.

So, I first consider the transition rates. Let $\underline{n} = (n_1, n_2)$ and $\underline{e_1} = (1,0), \underline{e_2} = (0,1)$.

Then, $q(\underline{n},\underline{n} + \underline{e_j}) = \lambda\frac{n_j + 1}{n_1 + n_2 + 2}$,

and $q(\underline{n},\underline{n} - \underline{e_j}) = \mu$.

Now usually, I know that an $M/M/\infty$ queue has invariant distribution which takes the form of a Poisson distribution with parameter $\rho$, and that $\pi(n_1,n_2) = \pi_1(n_1)\pi_2(n_2)$, however due to the balking that isn't the case here.

Any guidance would be appreciated, much thanks.

1

There are 1 best solutions below

0
On BEST ANSWER

Since you are only asked to verify that the equilibrium distribution is as given, you can just write down the balance equations for each state $(n_1,n_2)$ and plug in the equilibrium probabilities and check that the equations are indeed satisfied. You must also check if the equilibrium probabilities can be normalized, i.e. check if

\begin{equation} \sum_{n_1,n_2 \ge 0} \pi(n_1,n_2) < \infty. \end{equation}