0
$\begingroup$

Example: (Ross, p.338 #48(a)). Consider an $n$ -server parallel queueing system where customers arrive according to a Poisson process with rate $\lambda,$ where the service times are exponential random variables with rate $\mu,$ and where any arrival finding all servers busy immediately departs without receiving any service. If an arrival finds all servers busy,

Question: find the expected number of busy servers found by the next arrival.

It is easy to show that if we had a single server, and at $t=0$, it was busy, than the probability that the customers find the server busy is $\frac{\lambda}{\mu + \lambda}$. Hence $$T_1 := \text{# of busy servers} = 1 * \frac{\lambda}{\mu + \lambda} + 0*(1-\frac{\lambda}{\mu + \lambda}) = \frac{\lambda}{\mu + \lambda}.$$

Given that we have $n$ out of $n$ busy servers at $t=0$ which are independent from each other, the expected number of busy servers should be $$n* \frac{\lambda}{\mu + \lambda},$$ but in the lecture notes where I got this example, the author first finds $T_1$, then considers a $k$-server system with all servers busy and separates the cases according to whether first a customers arrives or at least one of servers finishes it's job. Then it tries to form a recursive relation between $T_i's$, and obtains quite a complex answers.

However, I cannot understand what is wrong with my answers. After all, we throw $n$ independent coins, each of which having probability $p$ to get heads, then we would expect to see $p*n$ heads. But, in this case, for some reason the auhors thinks that this logic does not work. Why?


I just made small simulation of the above system, with $n = 10$ and #of simulations = 100000. As far as numbers tell, the expected number of busy servers found by the customers is quite close to

$$\sum_{i=1}^n \frac{u_i}{u_i+\lambda}.$$ and this is the answer the author gives

enter image description here

$\endgroup$
1
$\begingroup$

Your approach and result are correct. You mention independence twice. The linearity of expectation does not require independence; your approach and result would be correct even without it. Since you didn’t give any details about the author’s more complicated solution, there’s not much more to say.

$\endgroup$
8
  • $\begingroup$ but the simulation does not agree with my answer? $\endgroup$ – Our Mar 7 '20 at 17:28
  • $\begingroup$ see the updated main post also. $\endgroup$ – Our Mar 7 '20 at 17:30
  • $\begingroup$ @onurcanbektas: I did see the update, but didn't understand it, since it contains variables $u_i$ that you didn't introduce. $\endgroup$ – joriki Mar 7 '20 at 21:54
  • 1
    $\begingroup$ @onurcanbektas: There's little point in repeating that without replying to my question above about the result of the simulation. In any case, what do you think is more likely – that there's a bug in your simulation (or perhaps you're misinterpreting its results), or that you, I, and Ross in the solution manual all made the same mistake? $\endgroup$ – joriki Mar 7 '20 at 22:45
  • 1
    $\begingroup$ There was indeed a bug in the code; thanks a lot for the help. $\endgroup$ – Our Mar 8 '20 at 11:22

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.