Reducing Overhead in Flow-Switched Networks: An Empirical Study of Web Tra c

| To eeciently transfer large amounts of diverse traac over high-speed links, modern integrated networks require more eecient packet-switching techniques that can capitalize on recent advances in switch hardware. Several promising approaches attempt to improve performance by creating dedicated \shortcut" connections for long-lived traac ows, at the expense of the network overhead for establishing and maintaining these shortcuts. The network can balance these cost-performance tradeoos through three tunable parameters: the granularity of ow end-point addresses , the timeout for grouping related packets into ows, and the trigger for migrating a long-lived ow to a shortcut connection. Drawing on a continuous one-week trace of Internet traf-c, we evaluate the processor and switch overheads for transferring HTTP server traac through a ow-switched network. In contrast to previous work, we focus on the full probability distributions of ow sizes and cost-performance metrics to highlight the subtle innuence of the HTTP protocol and user behavior on the performance of ow switching. We nd that moderate levels of aggregation and triggering yield signiicant reductions in overhead with a negligible reduction in performance. The traac characterization results further suggest schemes for limiting the shortcut setup rate and the number of simultaneous shortcuts by temporarily delaying the creation of shortcuts during peak load, and by aggregating related packets that share a portion of their routes through the network.