GREEN is a stateless queue-management algorithm that removes TCP's bias against connections with longer round-trip times, resulting in a high degree of fairness while maintaining high link utilization, low packet loss, and short queue sizes. GREEN applies the knowledge of the steady-state behavior of TCP connections to drop packets proactively, thus preventing long-lived TCP flows from ever inducing congestion. It also prevents shorter round-trip time (RTT) flows from grabbing more than their fair share of bandwidth. Consequently, GREEN ensures much higher fairness between flows than other queue-management schemes. Initial work on GREEN evaluated the performance of an ideal GREEN router, which was assumed to have global knowledge of each individual flow's RTT. Furthermore, the algorithm was susceptible to severe under-utilization of the link bandwidth in the presence of short-lived or low-bandwidth flows. To address these shortcomings, we present an incentive-based solution where flows benefit by embedding their current RTT estimates in the TCP header. We then present results for an IDMaps-based solution where the router estimates a flow's RTT by performing lookups in a local IDMaps database. Simulation results demonstrate that our algorithm is resilient for both long-lived as well as short-lived and low-bandwidth connections.
展开▼
