TBLB algorithm for servicing real-time multimedia traffic streams

In this paper, we propose to use a simple packet servicing algorithm suitable for servicing bursty real-time multimedia traffic streams in packet-switched networks. These real-time multimedia services may include packetized voice and videoconference/playback. The servicing mechanism is an enhancement of the token bank leaky bucket (TBLB) scheme we proposed previously. This new algorithm combines both the servicing and the policing functions, and its performance in accommodating bursty real-time traffic is evaluated by computer simulations. We show that the quality of service (QoS) performance (mean delay and jitter) of TBLB exceeds that of the leaky-bucket constrained generalized processor sharing (GPS). Although GPS has been proven to give bounded delay to a leaky-bucket constrained traffic stream and ensure instantaneous fair allocation of bandwidth, the average delay is often quite large. Also, fairness is not a guarantee of QoS, and is not perceived by users directly. Another property that is often neglected in the analysis of schedulers (but very important to user QoS) is the sensitivity of QoS to deviations of traffic streams from their specified traffic descriptors. We present results to show that our proposed method is relatively robust to such deviations