Applying statistical process control to the adaptive rate control problem

摘要

Abstract: Due to the heterogeneity and shared resource nature oftoday's computer network environments, the end-to-enddelivery of multimedia requires adaptive mechanisms tobe effective. We present a framework for the adaptivestreaming of heterogeneous media. We introduce theapplication of online statistical process control (SPC)to the problem of dynamic rate control. In SPC, thegoal is to establish (and preserve) a state ofstatistical quality control (i.e., controlledvariability around a target mean) over a process. Weconsider the end-to-end streaming of multimedia contentover the internet as the process to be controlled.First, at each client, we measure process performanceand apply statistical quality control (SQC) withrespect to application-level requirements. Then, weguide an adaptive rate control (ARC) problem at theserver based on the statistical significance of trendsand departures on these measurements. We show thisscheme facilitates handling of heterogeneous media.Last, because SPC is designed to monitor long-termprocess performance, we show that our online SPC schemecould be used to adapt to various degrees of long-term(network) variability (i.e., statistically significantprocess shifts as opposed to short-term randomfluctuations). We develop several examples and analyzeits statistical behavior and guarantees. !22