Credit-based low latency packet scheduling algorithm for real-time applications

Abstract

The number of real-time applications are growing dramatically at an unprecedented rate. Efficient bandwidth allocations and strict delay requirements are indispensable for real-time flows such as audio streaming. Unfortunately, most of the well-known packet scheduling algorithms like Weighted Fair Queueing (WFQ) and Start-time Fair Queueing (SFQ) can not guarantee low and stable latencies for real-time flows due to the `unsteady queueing delay problem' that we will explain later. Only a few scheduling algorithms address this problem in recent years like Low Latency Queueing (LLQ) and Low Latency and Efficient Packet Scheduling algorithm (LLEPS). However, LLQ has unfair and starvation problems. The queueing delay of LLEPS is not steady enough in existing networks. Thus, in this paper, we propose a novel credit-based low latency packet scheduling algorithm (CBLLPS) that is well-suited for real-time applications. The performance of CBLLPS is validated by ns-2 simulations. The simulation results show that CBLLPS guarantees low and steady queueing delay for real-time flows and outperforms any of the existing scheduling algorithms in terms of queueing delay. Moreover, the functionalities of bandwidth sharing and bandwidth reservation of this new scheduling mechanism are still preserved.