No abstract

Abstract. DataRouter forwards network layer packets using application layer tags, without requiring per-hop termination of transport protocols and the consequent reimplementation of transport services in the application. DataRouter provides network delivery based on pattern matching and string replacement. It combines a byte string as a loose source route IP option tag and regular expression routing entries to provide a new network service. DataRouter tags have a variety of forms, including fixed-length with exact matches for distributed hash tables and variable-length with regular expression matches for URL redirection. Tagged IPv6 packets traverse non-DataRouter routers transparently. On a platform forwarding IPv4 packets at 310K packets/sec., an unoptimized FreeBSD IPv4 DataRouter forwards hash-match packets at up to 270K packets/sec. (87% of max.) and pattern-match packets 155K packets/sec. (50% of max.). DataRouter thus provides a viable, higher-performance alternative to application-layer implementation of forwarding, in a generic service more interoperable with existing network and transport protocols.

No abstract

No abstract

Data access costs contribute significantly to the execution time of applications with complex data structures. As the latency of memory accesses becomes high relative to processor cycle times, application performance is increasingly limited by memory performance. In some situations it may be reasonable to trade increased computation costs for reduced memory costs. The contributions of this paper are three-fold: we provide a detailed analysis of the memory performance of a set of seven, memory-intensive benchmarks; we describe Computation Regrouping, a general, source-level approach to improving the overall performance of these applications by improving temporal locality to reduce cache and TLB miss ratios (and thus memory stall times); and we demonstrate significant performance improvements from applying Computation Regrouping to our suite of seven benchmarks. With Computation Regrouping, we observe an average speedup of 1.97, with individual speedups ranging from 1.26 to 3.03. Most of this improvement comes from eliminating memory stall time.