Internet traffic characterization

摘要: Traffic statistics normally collected during day-to-day operation of wide-area datagram networks are frequently insufficient for researchers to use in studying the workloads and performance these realistic environments. As become more ubiquitous service expectations rise, current methods collecting data will even less suitable. We examine ways improve techniques collection so that resulting enable researchers, indeed providers themselves, develop accurate Internet traffic models. We first provide a taxonomy characterization tasks. then operationally characterize T1 T3 NSFNET backbones. Because infrastructure is oriented toward either short term operational requirements or periodic simplistic reports funding agencies, this often not conducive assessing network workload performance; we evaluate what extent they useful tasks taxonomy, propose improvements architectures, with particular application backbone. include an investigation effects sampling high-speed environment. In second part thesis focus on items outlined using statistics. These mostly involve short-term aspects flows, which fail expose. general methodology flow profiles their impact aggregate workload. Our profiling flows differs from many previous studies have concentrated end-point definitions defined by TCP connections SYN FIN control mechanism. IP layer define based satisfying various temporal spatial locality conditions, as observed at internal points network. parameter space concentrate metrics characterizing both individual flow. Metrics include: volume packets bytes per flow, duration. characteristics perspective, counts number active, new, timed out time interval; interarrival arrival processes; metrics. Applying our measurements yields significant observations infrastructure, implications routers hotspots, specialized flow-based routing algorithms, future usage-based accounting requirements, prioritization. Finally, discuss trends affect how collect future. Improvements collection, such support assessment, help networking activities along horizons, defining quality patterns long-term capacity planning. offer unique combination research perspectives, allowing us reduce gaps among (1) need; (2) can provide; (3) analysis requires.