0
$\begingroup$

Let us suppose a scenario with two clients, $a$ and $b$, each one generating load at rate $\lambda_a$ and $\lambda_b$, respectively. The server receives the requests from both clients.

What will be the arrival rate at server?

(consider inter-arrival times independently generated by an exponential distribution)

$\endgroup$
  • 2
    $\begingroup$ The answer depends on whether the client request generation processes are independent processes or dependent processes. In the former case, the answer is $\lambda_a+\lambda_b$; in the latter, it is not. $\endgroup$ – Dilip Sarwate Aug 31 '15 at 14:25
  • $\begingroup$ @DilipSarwate, consider they are independent processes. Could you give an example of why the answer, then, is the sum of the rates? (thank you) $\endgroup$ – Lourenco Aug 31 '15 at 15:52
3
$\begingroup$

Let $A$ and $B$ denote the number of requests in $(0,T]$ from clients $a$ and $b$ relatively. Then, $A$ and $B$ are Poisson random variables with parameters $\lambda_aT$ and $\lambda_bT$ respectively, and you have declared them to be independent random variables. Thus, $C = A+B$, the total number of requests received during $(0,T]$ is a Poisson random variable also with parameter $(\lambda_a+\lambda_b)T$. This is because for any $N \geq 0$, \begin{align} P\{C = N\} &= P\{A+B = N\}\\ &= \sum_{n=0}^N P\{A= n, B = N-n\}\\ &= \sum_{n=0}^N P\{A= n\}P\{B = N-n\}&{\scriptstyle{\text{because $A$ and $B$ are independent random variables}}}\\ &= \sum_{n=0}^N \exp(-\lambda_aT)\frac{(\lambda_aT)^n}{n!} \exp(-\lambda_bT)\frac{(\lambda_bT)^{N-n}}{(N-n)!}\\ &= \exp(-(\lambda_a+\lambda_b)T)\sum_{n=0}^N \frac{(\lambda_aT)^n}{n!}\frac{(\lambda_bT)^{N-n}}{(N-n)!}\\ &= \frac{\exp(-(\lambda_a+\lambda_b)T)}{N!} \sum_{n=0}^N \binom{N}{n}(\lambda_aT)^n(\lambda_bT)^{N-n}\\ &= \exp(-(\lambda_a+\lambda_b)T)\frac{((\lambda_a+\lambda_b)T)^N}{N!} \end{align} When $A$ and $B$ are not independent, this does not work because the step above where independence was used is no longer valid.

$\endgroup$
  • $\begingroup$ Supposing $(0,T]$ is the time in which the samples happen, why let the Poisson random variables' parameters be $\lambda_a T$ and $\lambda_b T$ instead of $\lambda_a$ and $\lambda_b$? $\endgroup$ – Lourenco Sep 1 '15 at 2:42
  • 1
    $\begingroup$ $\lambda_a$ is the request rate, measured in number of requests per unit time. So, the average number of arrivals in $(0,T]$ is $\lambda_aT$, not just $\lambda_a$. $\endgroup$ – Dilip Sarwate Sep 1 '15 at 2:45
  • $\begingroup$ Yes, I got it. For instance, I also expand this question here. Could you comment there also? (thank you very much) $\endgroup$ – Lourenco Sep 1 '15 at 2:48
  • $\begingroup$ Considering the same proof you provided, the case of $\lambda_a = \lambda_b$ still valid, isn't it? (sorry for the naive confirmation) $\endgroup$ – Lourenco Sep 1 '15 at 18:08

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.