QoS provision with EDF scheduling, stochastic burstiness and stochastic guarantees

Abstract

Many of the typical network input processes can be characterized only stochastically. Consequently, only stochastic QOS guarantees can be provided to such traffic profiles. Such guarantees are usually sufficient, in view of the elasticity of typical (QoS demanding) applications, which translates into looser QoS requirements. Moreover, with stochastic guarantees, better network utilization can be achieved. Accordingly, this study considers QoS provision schemes for connections with stochastic traffic profiles and stochastic QoS requirements. We concentrate on the class of Rate-Controlled Earliest Deadlien First (RC-EDF) scheduling discipliens, which have several well known advantages, in particular simplicity of implementation and flexibility. Assuming the Exponentially Bounded Burstiness (EBB) traffic model, we establish results that extend the deterministic study of RC-EDF, both for a single server in isolation and for networks of servers. For a single traffic shaper followed by an EDF scheduler. we establish stochastic bounds on the distribution of the delay for each session. In the general (multi-hop) setting, we first establish stochastic bounds on the distribution of the end-to-end delay for traffic shaper elements in series; then, we establish stochastic bounds for RC-EDF networks. The application of these bounds is illustrated through an example, in which our bounds practically outperform the corresponding bounds for RPPS-PGS networks with EBB traffic.