On the Effectiveness of the Longest-Packet-In Selective Packet Discarding Policy

In this paper, we introduce a novel packet discarding policy called Longest-Packet-In (LPI) which maximizes the cell goodput, the ratio of outgoing good cells to the total number of cells that arrive at a queue. We compare LPI to: Tail Drop, Early Packet Discard, Early Packet Discard with Hysteresis and Fair Early Packet Discard with Hysteresis by investigating the trade-off between maximizing the cell goodput and maximizing the packet goodput. We show that LPI not only provides the highest cell goodput, but also produces the highest packet goodput for a high ratio of the mean packet size to the buffer size and under a short-range dependent process. We evaluate the impact of network load, mean packet size, packet distribution and the Hurst parameter into both the cell goodput and into the packet goodput. Furthermore, we extend our analysis to networks which carry sources with distinct mean packet length.