This invention relates to the field of
queue management technology for
Ethernet switching chips, specifically to a method for managing the probabilistic random drop of packets in a
queue, applied to multi-service
queue management in
Ethernet switching chips. The method configures drop thresholds, sampling periods, prediction step sizes, congestion change rate thresholds, and target drop ratios for each service queue; calculates the congestion change rate and predicted queue length based on the real-
time average queue length and its historical sequence, and determines the direction of threshold adjustment accordingly; determines the adjustment step value based on the deviation between the actual drop ratio and the target drop ratio, and updates the minimum and maximum drop thresholds. When packets are enqueued, a pseudo-random number is generated using a
linear feedback shift register, and combined with the congestion change rate to generate a dynamic comparison threshold to complete the random drop judgment. This invention can improve queue congestion response speed and reduce hardware storage
resource consumption.