Chapter 12

Delay Systems

In this chapter we consider trac to a system with n identical servers, full accessibility, and
an innite number of waiting positions. When all n servers are busy an arriving customer
joins a queue and waits until a server becomes idle. No customers can be in queue when a
server is idle (full accessibility).

We consider the same two trac cases as in Chaps. 7 & 8.

1. Poisson arrival process (an unlimited number of sources) and exponentially distributed
service times (PCT-I). This is the most important queueing system called Erlang s
delay system. Using the notation introduced later in Sec. 13.1, this system is denoted
as M/M/n. In this system the carried trac will be equal to the oered trac as no
customers are blocked. The probability of a positive waiting time, mean queue lengths,
mean waiting times, carried trac per channel, and improvement functions will be dealt
with in Sec. 12.2. In Sec. 12.3 Moe s principleis applied for optimizing the system.
The waiting time distribution is calculated for three basic service disciplines: First{
Come First{Served (FCFS, Sec. 12.4), Last{Come First-Served (LCFS, Sec. ??), and
Service In Random Order (SIRO, Sec. ??).

2. A limited number of sources and exponentially distributed service times (PCT-II ). This
is Palm s machine repair model(the machine interference problem) which is dealt with
in Sec. 12.5. This model has been widely applied for dimensioning computer systems,
terminal systems, and e.g. exible manufacturing system (FMS). Palm s machine repair
model is optimised in Sec. 12.6.
A general birth & death process, which includes many of the models considered, is presented
in Sec. ?? to illustrate the general priciples. We follow the cookbook given in Sec. 7.4 to set up
the state transition diagram andn d state probabilities under the assumption of statistical
equilibrium. From the state probabilities wen d mean queue lengths, and by means of Little s
law we obtain the mean waiting times. As special cases we get Erlang s delay system withn
servers and a limited number k n, (nk ), of queueing positions (nite buer) ( M/M/n,k).

191
196

Figure 12.2: Erlang s C{formula for the delay system M/M/n. The probability of waiting
E2;n (A) for a positive waiting time is shown as a function of the oered trac A for dierent
values of the number of servers n.