Adaptive traffic-shaping

Abstract

Bounding the transmission latency of a priority flow with a standard traffic-shaper is typically accomplished by applying worst-case analysis. The network designer-often having limited detailed knowledge of the application data, its host protocol configuration, and full network path overhead-rightfully assumes maximum protocol overhead, largest possible data load, zero-gain compression, etc. The consequence of this practice is a catastrophic impact on traffic that competes with the priority flow under anemic network conditions. The work presented here provides an alternative to static trafficshaping that guarantees bandwidth availability to priority flows while minimizing their impact on competing traffic. An adaptive traffic-shaper implements algorithms that perform online analysis of the stochastic character of network traffic and "learns" to optimize total quality of service while respecting transmission constraints on priority flows.