Effective bandwidth vectors for multiclass traffic multiplexed in partitioned buffer

We consider a traffic model where a single source generates traffic having J (J⩾2) quality of service (QoS) classes, QoS in this case described by a cell loss probability objective ε_j for QoS class j. We assume that ε₁⩾ε₂ ⩾...⩾ε_J, in other words, class J has the most stringent QoS requirements and class 1 the least. The traffic from K such independent heterogeneous Markov-modulated fluid sources is multiplexed into a single buffer of size B. There are J-1 thresholds {B _j, 1⩽j⩽J-1} such that 0<B₁<B₂ <...<B_J-1<B. Let B₀=0 and B_J =B. If the buffer content is in (B_j-1, B_j), 1⩽j⩽J only traffic of class index J or above is accepted and all other traffic is rejected. For this system of K sources we define an effective bandwidth vector of size J such that QoS requirements for all classes are satisfied if each component of the vector is less than the channel capacity. We propose several bandwidth vectors that can be computed for each source separately. Numerical studies are reported on the efficacy of these bandwidth vectors