5.2 Multiple Server Queues.

The previous section considered single server queues. In this section we extend the analysis to queues
with m servers. This is often called an M/M/m queue. We let

 = arrival rate of new customers

 = service rate = rate at which each server serves customers

m = number of servers

N(t) = number of customers waiting at time t including the one being served

Thus, 1/ is the average time between customer arrivals and 1/  is the average service time of a single
custormer. t measures time relative to some starting time. We regard N(t) as a random variable, so N(t) is
a Markov process. The states are the non-negative integers n = 0, 1. 2, …

Example 1. A company has a telephone number where you can call to place an order. It has two people
(the servers) answering calls. If they are busy with one order, an answering machine places further calls
on hold and they are answered in the order they are received when the servers finish with customers.
Assume the time between new calls is an exponential random variable with mean three minutes, so  =
1/3. Also assume the time each server takes to handle a call is an exponential random variable with mean
two minutes, so  = ½. We assume all times are independent of all the other times. The transition
diagram is
       
0 1 2 3 n-1 n n+1
 2 2 2 2 2 2 2

Note that all the left going arrows are labeled with 2  except the one going out of state 1. This is because
for all the states n with n  2 there are two customers being served each with completion rate . However
for the state n = 1 there is only one customer being served with completion rate .

Let n be the steady state probability of being in state n, i.e. of there being n customers in the queue
including the one being served. For the state 0 the balance equation is

transition rate out of state 0 = transition rate into state 0  0 = 1  1 = 0

For the state 1 the balance equation is

transition rate out of state 1 = transition rate into state 1  0 + 22 = ( + )1

Using 0 = 1, this becomes

5.2 - 1
 22 = 1  2 = 1 = 1 = 220

where

 =

For the state 2 the balance equation is

transition rate out of state 2 = transition rate into state 2  1 + 23 = ( + 2)2

Using 1 = 22, this becomes

23 = 2  3 = 2 = 230

Continuing in this fashion we see that

n = 2n0

Now we use the fact that the sum of the n must be one.

1 = 0 + 1 + 2 + 3 +  + n + 

= 0 + 20 + 220 + 230 +  + 2n0 + 

= 0 (1 + 2 + 22 + 23 +  + 2n +  )

= 0 [2(1 +  + 2 + 3 +  + n +  ) – 1]

= 0 [2 – 1] = 0

So

0 = (as compared with 0 = 1 -  with one server)

n = 2 n for n  1 (as compared with n = (1 - )n with one server)

In Example 1 we have

 = = = as compared with  = with one server

0 = = = as compared with 0 = with one server

1 = 2  = as compared with 1 = with one server

n = 2 n = n
as compared with n = n
with one server

Now let's compute the average queue length and waiting time

5.2 - 2
 Lq = average queue length (not including those being served)

= 00 + 01 + 02 + 13 + 24 +  + (n – 2)n + 

= 2 + 222 + 332 +  + (n – 2)n-22 + 

= 2 [1 + 22 + 32 +  + (n – 2)n-1 +  ]

= =

Wq = average waiting time (not including the service time)

= = =

W = average time in queue including the service time

= Wq + = +

= = (W = with one server)

L = average number in queue including the one's being served

= W = = (L = with one server)

In Example 1 we have

W = = = min = 2.25 min (W = 6 min for one server)

L = = = customer (L = 2 for one server)

If we put a cost of c = 30¢ / customer - minute for customers waiting time, then

Ctime = cL = (30¢ / customer - minute) = 22.5¢ / min = $13.50 / hr

as compared with $36.00 / hr with one server. So if it costs $15.00 / hr to hire another server, then from
one point of view it might be worthwhile. The cost per customer is

Ccustomer = cW = (30¢ / customer - minute) = 67.5¢ / customer

as compared with $1.80 /customer with one server.

If there are than two customers the analysis is similar, but a liitle more complicated. After some work one
obtains the following formulas for m servers

0 =

n =

5.2 - 3
Lq =

Wq =

W = Wq +

L = W = Lq + c

5.2 - 4