Information packets arrive at a server with a poisson process having rate $\lambda = 2$ per hour.

The server processing time for a packet follows the distribution : $f(x) = 1, 0\leq x\leq1$

The status of the server is busy if it is processing a packet, otherwise it is waiting. If a packet arrives while the server is busy, that packet is lost.

Let $t_B$ denote the length of one busy period, $t_I$ denote the length of one idle period. Find the distrubutions of $t_B$ & $t_I$.

Let $N$ denote the total number of lost packets, $T_B$ denote the total busy time of the server up until time $T=10$ hours.

Show that : $$ \begin{align} E[N] = \lambda E[T_B] \end{align} $$

My attempt :

$t_B$ is simply the time taken for the server to process a packet, so its distribution is also $f(x) = 1, 0 \leq x \leq 1$

$t_I$ is the inter-arrival time between two packets which follows an $Exp(\lambda = 2)$ distribution, using the relation between poisson & exponential rv's.

In trying to prove the equation, this is what I have so far :

N = total no. of lost packets = no. of packets arriving during $T_B$(while the server is busy)

$$ \begin{align} E[N] &= E[\textrm{No. of arrivals during }T_B] \\ &=E[\textrm{No. of arrivals per unit time}]\cdot E[T_B] \\ &=\lambda E[T_B]~\textrm{(Shown)} \end{align} $$

Am I going about this the right way? If so is there a proper way to express this in math notation? Thank you!


1 Answer 1


Because the packets arrive as a Poisson process, with a given $T_{B}$, the number of missing packets $N$ also follows a Poisson distribution, $$ p(N \vert T_B) = {\rm Poisson}(\lambda T_B)(N). $$ The expected value of $N$, ${\rm E}[N]$ is then given by $$ \begin{aligned} {\rm E}[N] &= \sum_{N=0}^{\infty} \ N p(N)\\ &= \sum_{N=0}^{\infty} N \int_0^\infty dT_B \ p(N, T_B)\\ &= \sum_{N=0}^{\infty} N \int_0^\infty dT_B \ p(N \vert T_B) p(T_B)\\ &= \int_0^\infty dT_B \ p(T_B) \sum_{N=0}^{\infty} N p(N \vert T_B)\\ &= \int_0^\infty dT_B \ p(T_B) {\rm E}[N \vert T_B]\\ &= \int_0^\infty dT_B \ p(T_B) \lambda T_B\\ &= \lambda \int_0^\infty dT_B \ T_B p(T_B)\\ &= \lambda {\rm E}[T_B]. \end{aligned} $$

  • $\begingroup$ Thank you for your answer! I think I understand. Could I check why $E(N|T_B) = \lambda T_B$? Since $p(N|T_B) = Poisson(\lambda T_B)(N)$ wouldn't $E(N|T_B) = \lambda T_B N$? Have I misunderstood something? $\endgroup$
    – C C
    Commented Apr 9, 2022 at 15:01
  • $\begingroup$ The notion of $E(N\vert T_B)$ is the expected value of $N$ given $T_B$. Because the distribution $N\vert T_B \sim {\rm Poisson}(\lambda T_B)$, the expected value is simply $\lambda T_B$ as it is the property of a Poisson distribution. $\endgroup$
    – Peter Pang
    Commented Apr 10, 2022 at 19:18
  • $\begingroup$ I understand. Thank you so much! :) $\endgroup$
    – C C
    Commented Apr 15, 2022 at 7:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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