Achieving fair and predictable service differentiation through traffic degradation policies

Recently a large number of approaches to service differentiation have appeared in the literature. Most of these approaches are based on the Differentiated Services (DiffServ) Architecture proposed by IETF and they have been evaluated in a number of independent studies. It is widely acknowledge that the Differentiated Services approach provides proper service differentiation in well-provisioned networks under normal traffic conditions. However DiffServ may fail to provide proper service differentiation in the presence of extreme network conditions such as congestion or in under-provisioned networks, resulting in unfair service degradation and unpredictable traffic behavior. We believe that the main reasons for the failure of service differentiation under extreme network conditions are static per-aggregate resource allocation and the lack of the specification of traffic behavior during the congestion. We propose to explicitly define traffic behavior through so called degradation policies. Degradation policies would specify how much of the network resources each traffic class could receive under different levels of congestion. In particular, the degradation policy would specify traffic behavior in terms of the performance parameters like delay, loss, or throughput. Such an approach to service differentiation allows dynamic allocation of traffic resources while maintaining fairness and predictability of the traffic behavior under all network conditions.