Probabilistic bandwidth reservation by resource priority multiplexing

Probabilistic bandwidth guarantees on variable bit-rate network flows offer significant improvements in throughput for only modest decreases in guarantee level. We present a probabilistic bandwidth reservation scheme for variable bit-rate flows called resource priority multiplexing (RPM). The RPM algorithm uses packet marking on edge routers and selective dropping on core routers to provide per-flow specification of probabilistic QoS with guarantees ranging from best-effort to hard reservation. RPM also supports multiple metrics for probabilistic level of service specification. All flow tables are managed only by the edge routers enhancing scalability of the approach. Unlike existing statistical multiplexing approaches, RPM uses a time-multiplexed priority tagging algorithm which provides higher control over flows guarantee levels. The algorithm has been implemented in a live system as part of the IP protocol stack and experimental results are presented.