The effects of active queue management and explicit congestion notification on web performance

We present an empirical study of the effects of active queue man- agement (AQM) and explicit congestion notification (ECN) on the distribution of response times experienced by a population of users browsing the Web. Three prominent AQM schemes are consid- ered: the Proportional Integral (PI) controller, the Random Expo- nential Marking (REM) controller, and Adaptive Random Early Detection (ARED). The effects of these AQM schemes were stud- ied with and without ECN. Our primary measure of performance is the end-to-end response time for HTTP request-response ex- changes. For this measure, our major results are:  If ECN is not supported, ARED operating in byte-mode was the best performing AQM scheme, providing better response time performance than drop-tail FIFO queuing at offered loads above 90% of link capacity. However, ARED operating in packet- mode (with or without ECN) was the worst performing scheme, performing worse than drop-tail FIFO queuing.  ECN support is beneficial to PI and REM. With ECN, PI and REM were the best performing overall schemes, providing sig- nificant response time improvement over ARED operating in byte-mode. In the case of REM, the benefit of ECN was dra- matic. Without ECN, response time performance with REM was worse than drop-tail FIFO queuing at all loads considered.  ECN was not beneficial to ARED. Under current ECN imple- mentation guidelines, ECN had no effect on ARED perform- ance. However, ARED performance with ECN improved sig- nificantly after reversing a guideline that was intended to police unresponsive flows. Nonetheless, overall, the best ARED per- formance was achieved without ECN.  Whether or not the improvement in response times with AQM is significant (when compared to drop-tail FIFO), depends heavily on the range of round-trip times (RTTs) experienced by flows. As the variation in flows' RTT increases, the impact of AQM and ECN on response-time performance is reduced. We conclude that AQM can improve application and network per- formance for Web or Web-like workloads. In particular, it appears likely that with AQM and ECN, provider links may be operated at near saturation levels without significant degradation in user- perceived performance.