This paper compares the performance of Tail Drop and three different flavors of the RED (Random Early Detection) queue management mechanism: RED with a standard parameter setting, RED with an optimized parameter setting based on a model of RED with TCP flows, and finally a version of RED with a smoother drop function called "gentle RED". Performance is evaluated under various load situations for FTP-like and Web-like flows, respectively. We use measurements and simulations to evaluate the performance of the queue management mechanisms and assess their impact on a set of operator oriented performance metrics. We find that in total (i) no performance improvements of RED compared to Tail Drop can be observed; (ii) fine tuning of RED parameters is not sufficient to cope with undesired RED behavior due to the variability in traffic load; (iii) gentle RED is capable of resolving some of the headaches on RED but not all.

The paper presents a technique for computing the individual throughputs and the average queue occupancy when multiple TCP connections share a single bottleneck buffer. The bottleneck buffer is assumed to perform congestion feedback via randomized packet marking or drops. We first present a fixed point-based analytical technique to compute the mean congestion window sizes, the mean queue occupancy and the individual throughputs when the TCP flows perform idealized congestion avoidance. We subsequently extend the technique to analyze the case where TCP flows perform generalized congestion avoidance and demonstrate the use of this technique under the Assured Service model, where each flow is assured a minimum traffic rate. Simulations are used to demonstrate the accuracy of this technique for relatively low values of packet dropping probability and a much wider range of packet marking probability.

Design and engineering of networks requires the consideration of many possible configurations (different network topologies, bandwidths, traffic and policies). Network engineers may use network simulation to evaluate changes in network configuration, but detailed, packet-level simulation of many alternatives would be extremely time consuming. This paper introduces the concept of scenario prefiltering ---rather than perform detailed simulation of each scenario, we propose to quickly evaluate (pre-filter) all scenarios in order to select only the relevant scenarios and discard those that are clearly too over- or under-provisioned. To rapidly evaluate scenarios, we have developed several new analytical techniques to quickly determine the steadystate behavior of the network with both bulk and short term TCP flows. These techniques apply to arbitrary topologies and routers that use both drop-tail and RED queuing policies. Since we are only interested in selecting the interesting scenarios for detailed simulation, the answers need only be approximate. However, we show that accuracy is typically within 10% of detailed simulation. More importantly, these techniques are 10--300 faster than detailed simulation, and, hence, pre-filtering is a promising technique to reduce the total simulation time when many scenarios must be considered.