Network fair bandwidth share using hash rate estimation

In this paper, we evaluate the effectiveness of the current active queue management (AQM) schemes, such as Random Early Detection (RED) and match-based CHOKe, over traffic with different packet-size flows and over traffic with mixed adaptive and nonadaptive flows. We find that these schemes allow unfair bandwidth sharing when flows with different packet sizes share a link or when numerous adaptive and nonadaptive flows coexist in a link. In particular, the match-based schemes present greater unfairness, especially when more nonadaptive flows exist in a shared link. We propose a novel AQM scheme, called Hash Rate Estimation (HRE). The proposed scheme, which works with a first-in-first-out queue, uses per-active-flow rate estimation to impose a fair share on each flow. We show that HRE provides better protection than does Fair Random Early Drop (FRED), which uses per-flow accounting to allocate each flow its fair share by a fixed dropping probability that depends on the flow's buffer use. HRE can support a large number of flows using hash function sorting. In addition, the queue length under HRE is bounded and the characteristics of traffic regulated by HRE are similar to those achieved by the per-flow fair queuing schemes.