A Machine receives two kinds of jobs following two Independent Poisson Processes $1$ and $2$ with arrival rates $\lambda_{1}$ and $\lambda_{2}$, respectively. The machine process the jobs in the following way:

  1. If the $1^{st}$ job is coming from process $1$, it will have $0$ wait, and process repeats.
  2. If the $1^{st}$ job is coming from process $2$, it will either wait for time $T$, OR another job from process $1$ before $T$ expires. Process repeats.

Focusing on Point $2$: What is the probability that $1^{st}$ job of process $2$ is served due to a job coming from process $1$ before $T$ expiry?

My answer is: $$P\ (1\ or \ more \ arrivals \ from\ Process\ 1 \ in [0-T]\ )=1-e^{-\lambda_{1}T}$$

I developed a simulator to check this probability value.

  1. When $\lambda_{1} >= \lambda_{2}$, both simulator and above expression results matches, and probability gets $1$ for higher rates.

  2. When $\lambda_{1} <$ $\lambda_{2}$, both simulator and the above expression results do not match for higher rates. Surprisingly, simulation results get constant to a value lower than 1, as $\lambda$ increases.

What would be the right expression for calculating this probability in case if $\lambda_{1} <$ $\lambda_{2}$?

  • $\begingroup$ So by point $2$ do you mean the time to serve a job from stream $2$ is $\min\{T, \tau\}$ where $\tau$ is the time until the next job from stream $1$? $\endgroup$ – Nap D. Lover Nov 2 '17 at 11:53
  • $\begingroup$ Can you show us what your simulation does? $\endgroup$ – Raskolnikov Nov 2 '17 at 12:21
  • 2
    $\begingroup$ Ah, but your comment to LoveTooNap29 changes everything. Of course the probability is not $1-exp{-\lambda_1 T}$ when you add a third condition. On top of that, when you increase the rate of the second process, you augment the chance that 2 will be processed because there are k jobs in wait. $\endgroup$ – Raskolnikov Nov 3 '17 at 4:36

What you need is $$\mathbb{P}(\tau<\mu)$$

that is, the probability that a job from process 1 arrives before $k$ jobs from process 2 arrive.

The time of arrival for process 1 is indeed exponentially distributed with parameter $\lambda_1$. However the time of arrival of $k$ processes is Erlang distributed, i.e. its density is

$$f_{\lambda_2,k}(t)=\frac{\lambda_2^k t^{k-1}e^{-\lambda_2 t}}{(k-1)!}$$

I worked out the end result, it is:

$$\frac{\Gamma(k;(\lambda_2) T)}{(k-1)!}-\left(\frac{\lambda_2}{\lambda_1+\lambda_2}\right)^k\frac{\Gamma(k;(\lambda_1+\lambda_2) T)}{(k-1)!}$$

in which $\Gamma(\alpha,x)$ is the incomplete Gamma function with parameter $\alpha$.

  • $\begingroup$ Thanks a lot, let me go through it to get the curves. $\endgroup$ – Hallian1990 Nov 5 '17 at 10:47
  • $\begingroup$ Now that I have been rethinking about the problem: do you need $\mathbb{P}(\tau<\mu)$ or $\mathbb{P}(\tau<\mu|\min(\tau,\mu)<T)$? This makes an important difference. I also think I inadvertantly computed $\mathbb{P}(\tau<\mu \text{ and } \min(\tau,\mu)<T)$ which is still something else. It would help if you showed me the code of your simulation. $\endgroup$ – Raskolnikov Nov 5 '17 at 10:56
  • $\begingroup$ You asked good point. Since we have three conditions for serving Stream $2$ i.e., due to $T$, $k$ stream 2 arrivals, or one stream $1$ arrival, the probability that, the Stream $2$ get service due to an incoming Stream $1$ arrival means the rest of the two conditions are not fulfilled. This is exactly how simulator conditions work. Therefore, this probability would be $Pr \ (to \ get \ stream \ 1 \ arrival \ before \ 'k' \ stream \ 2 \ arrivals, \ within \ time \ T ) $. So, it will be the first expression of your comment, $p(\tau< \mu | min(\tau,\mu)<T)$ $\endgroup$ – Hallian1990 Nov 5 '17 at 11:15

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.