TCP download performance in dense WiFi scenarios

摘要: How does a dense WiFi network perform, specifically for the common case of TCP download? While empirical answer to this question is ‘poor’, analysis and experimentation in prior work has indicated that clocks itself quite well, avoiding contention-driven overload settings. This paper focuses on measurements from real-life use scenario: classroom where several students download quizzes instruction material. We find performance poor, contrary suggested by work. Through careful analysis, we explain complex interaction various phenomena which leads poor performance. Specifically, observe small amount upload traffic generated when downloading data upsets clocking, increases contention channel. Further, losses lead vicious cycle with autorate adaptation TCP's timeout mechanism. To reduce channel improve performance, propose modification AP scheduling policy large downloads. Our solution, WiFiRR, picks only subset clients be served during any instant, varies set “active” periodically round-robin fashion over all ensure no client starves. By reducing number contending nodes at point time, WiFiRR improves time flows 3.2× simulation our scenario.