Performance space of a GI/G/1 queueing system under a percentile goal criterion

The delay performance of a GT/G/1 queueing system with a single server and multiple classes is studied under a new criterion, namely the percentile goal criterion. We assume independent and identically distributed general interarrival times and general service times for each class of packets. The percentile goal criterion is characterized by a prespecified waiting time objective and the goal is to guarantee that a given fraction, or percentile, of packets from a class will be served with their waiting times less than the specified waiting time objective. We first characterize the performance space of such queueing systems. Then we describe a mixing strategy which achieves a given percentile goal. Later, we describe an adaptive algorithm which solves a relevant maxmin problem