Sammendrag
The thesis ``Ytelse og skalerbarhet i distribuerte
telekommunikasjonssystemer'' contains an analysis of the behavior of six different
algorithms. The algorithms are simple, and they schedule customers from multiple queues
in a Call-center. A simulator is implemented i Java to
measure the customers waiting time and the non-operating time of the
agents. The algorithms are compared to each other so that the best
method can be chosen. An example of a good method is one that gives
the customers short waiting time.
Two different cases was tested. CASE 1 was set up to handle two
queues with the same arrival rate. One type of agents served both
of the queues, and the average serving rate was estimated to
the same value for all of the arriving customers. In this case,
the algorithms produced almost the same result when the arrival rate, serving
rate or the number of agents in the system was changed.
CASE 2 was configured with three different agent types and three
queues. Each queue had their own arrival rate, and the agent groups had
different serving rate, they also served different sets of
queues. In this case the result from the algorithms differ. The
customers waiting time and the agents non-operating time was dependent
on the choice of algorithm.
When the arrival rate was changed, the static algorithm type gave
increased waiting time for the customers in the queue where the
arrival rate increased. The customers in the other queues did just get a
insignificance increase in their waiting time. For dynamic queue scheduling
the result was a little different. When the arrival rate increased in
one queue, the waiting time increased in all queues in the
system. This is because these algorithms analyse each queue load
before an customer is picked from a queue.
The agents non-operating time decreased more when a dynamic
algorithm was used to schedule the next customer. The reduction of the
non-operating time and the degree of increase in waiting time shows that
dynamic algorithms should be chosen if the system is going to
have high performance and scalability.