• Summary
• Related Documents
  • Active Bibliography
  • Co-citation
• Version History

Abstract:

Several recent proposals have been made for sharing congestion information across concurrent ows between end-systems, where the proposed granularity for sharing has ranged from all ows to a common host, to all hosts on a shared LAN. This paper addresses the problem of false sharing caused by these proposals: two or more ows sharing congestion state may in fact not share the same bottleneck. We characterize the origins of false sharing into two distinct cases: (i) networks with QoS enhancements such as dierentiated services, where a ow classier segregates ows into dierent queues, and (ii) networks with path diversity where dierent ows to the same destination address are routed dierently, a situation that occurs in dispersity routing, load-balancing, and with network address translators (NATs). We evaluate the impact of false sharing on ow performance and consider whether it might cause a bottleneck link to become persistently overloaded. We then consider how false sharing can be detected by a sender and how dierent metrics (loss rate, delay distribution, and reordering) compare for this purpose. Finally, we consider the issue of how a sender must respond when it detects false sharing. Our simulation results show that persistent overload can be avoided with window-based congestion control even for extreme false sharing, but higher bandwidth ows run at a slower rate. We nd that delay and reordering statistics can be used to develop robust detectors of false sharing and are superior to those based on loss patterns. We also nd, somewhat surprisingly, that it is markedly easier to detect and react to false sharing than it is to start by isolating ows and merging their congestion state together afterwards. 1

Citations