In order to support Quality of Service, different service disciplines have been developed for packet scheduling. One of the most popular choices is the Generalized Processor Sharing (GPS). For such a discipline delay or probability of delay are bounded for a leaky-bucket or EBB envelopes.

In this work, we propose a routing framework that is specifically tailored to the GPS discipline. First,we present an algorithm that quickly generates several suitable paths and increases the chances that the QoS is satisfied.

The algorithm is used in conjunction with deterministic and stochastic models of GPS to select the best path for a new data flow. Next, we investigate the impact on QoS of changing the routing decision based on probabilistic and historic data. For this, several decision policies are presented and explored. Finally, we assess the QoS based on different admission control policies.