| Internet flow characterization: Adaptive timeout strategy and statistical modeling (2001) | |||||||||||||||||
Abstract | |||||||||||||||||
| Abstract | There is a growing eort on understanding Internet trac dynamics via the abstraction of ows from packets. We rst present an adaptive ow timeout strategy for identifying and maintaining ow state information. We demonstrate that, via an extensive analysis of a large number of traces collected at three dierent sites, this adaptive strategy has the potential to achieve signicant performance advantages over widely used xed timeout scheme. These advantages translate to faster detection of the end of ow, more accurate recognition of long-lived ows and substantial reduction in hardware resources. Second, we present a highly versatile and parameterized statistical modeling framework for characterizing a broad spectrum of ow dynamics. This framework identies several components key to the faithful description of ow dynamics in a statistical sense. Based on the analysis of a couple of very large traces, we nd that the proposed modeling framework are capable of capturing some of the essential characteristics of ow dynamics. We also identify that some of the widely used assumptions about ow modeling, such as Poisson ow arrivals and uniform/Poisson packet arrivals within ows underestimate the burstiness of ow dynamics. Instead, we nd that ow arrivals appear to follow mono-fractal characteristics over time scales greater than 100 msec, and intra-ow packet dynamics exhibit an array of dierent statistical behaviors, including non-fractal, mono-fractal, and multifractal patterns. While these deviations from common modeling assumptions make faithful ow dynamics modeling a challenge, the proposed modeling framework paves the way for accurate ow simulation and provides plausible causes for complex packet-level dynamics such as multi-fractals. | |||||||||||||||||
Details der Publikation | |||||||||||||||||
| |||||||||||||||||