Improving the network performance using prediction based longest queue first (PLQF) scheduling algorithm

Abstract

Traffic today is broadband and bursty by nature. Extended overloads contribute to network congestion and increase the probability of buffer overruns and packet discarding. How to reduce the CLR (cell loss ratio) is an important issue that must eventually be confronted. In this work, we propose the prediction-based longest queue first (PLQF) scheduling algorithm. During scheduling, the scheduler considers not only the queue length, but also the incoming traffic. Based on that information, the resource is allocated to the user with the highest probability of overflow. The scheduler can adjust the queue lengths in advance to cater to the incoming traffic, targeting to minimize CLR and improve the buffer utilisation. Theoretical analysis shows that the PLQF algorithm can lead to a lower CLR than the conventional LQF algorithm, especially for bursty traffic. Simulation results prove this to be true. Compared with conventional algorithms, the PLQF algorithm can reduce CLR by 10%-60%. Therefore, the PLQF algorithm is promising as an efficient scheduling algorithm.