The computer connection to INTERNET is provided by firms known as INTERNET access providers, which can handle several kinds of users by offering different forms to of physical connection. The simplest one is when the user connects to a provider's attending channel by an ordinary telephone line. Finding an available channel may not be an easy task, especially during peak hours. This results in a problem for the providers, which is to determine the optimal relationship between the number of users and the number of attending channels. To solve this problem, a provider needs to analyze the tradeoff between investing in capacity (number of channels) and satisfying the targeted service level to the users (probability of finding an available channel). The objective of this paper is to model this tradeoff by means of queueing theory. The methodology involves basically three steps: (i) analyze user arrival (calling) and service (attending) processes in chosen periods, (ii) select an appropriate queueing model under some simplifying assumptions, and (iii) generate tradeoff curves among system performance measures, in particular, between the probability of finding an available channel and the number of channels or the mean user arrival rate. To illustrate the application of this methodology, some results deriving from a case study performed in an access provider of São Paulo state, with Poisson arrival process but non-exponential service times, are presented. Some perspectives for future research are pointed out, such as grouping users into different classes as a function of their arrival and service behavior, and analyzing capacity partitioning as a function of these classes.
INTERNET provider; queueing theory; service level
