Bias Against Bursty Traffic
Assume bursts (batches) of B packets arriving as Poisson
Lower bound on difference between smooth traffic and more bursty (e.g. LRD) traffic
Use PASTA (Poisson Arrival See Time Averages)
Compute drop probability for TD
Assume instantaneous queue size measurements (weight = 1)
Approximation 1: same drop probability used on all packets in same burst
Accurate for small min_thresh, high maxP, B << buffer size K
Drop probability is always higher with RED than TD
For high load, loss probability for RED almost equals that of M/M/1/K Tail Drop