Performance analysis of an ATM network with multimedia traffic : a simulation study
Traffic and congestion control are important in enabling ATM networks to maintain the Quality of Service (QoS) required by end users. A Call Admission Control (CAC) strategy ensures that the network has sufficient resources available at the start of each call, but this does not prevent a traffic source from violating the negotiated contract. A policing strategy (User Parameter Control (UPC)) is also required to enforce the negotiated rates for a particular connection and to protect conforming users from network overload. The aim of this work is to investigate traffic policing and bandwidth management at the User to Network Interface (UNI). A policing function is proposed which is based on the leaky bucket (LB) which offers improved performance for both real time (RT) traffic such as speech and video and non-real time (non-RT) traffic, mainly data by taking into account the QoS requirements. A video cell in violation of the negotiated bit rate causes the remainder of the slice to be discarded. This 'tail clipping' provides protection for the decoder from damaged video slices. Speech cells are coded using a frequency domain coder, which places the most significant bits of a double speech sample into a high priority cell and the least significant bits into a high priority cell. In the case of congestion, the low priority cell can be discarded with little impact on the intelligibility of the received speech. However, data cells require loss-free delivery and are buffered rather than being discarded or tagged for subsequent deletion. This triple strategy is termed the super leaky bucket (SLB). Separate queues for RT and non-RT traffic, are also proposed at the multiplexer, with non pre-emptive priority service for RT traffic if the queue exceeds a predetermined threshold. If the RT queue continues to grow beyond a second threshold, then all low priority cells (mainly speech) are discarded. This scheme protects non-RT traffic from being tagged and subsequently discarded, by queueing the cells and also by throttling back non-RT sources during periods of congestion. It also prevents the RT cells from being delayed excessively in the multiplexer queue. A simulation model has been designed and implemented to test the proposal. Realistic sources have been incorporated into the model to simulate the types of traffic which could be expected on an ATM network. The results show that the S-LB outperforms the standard LB for video cells. The number of cells discarded and the resulting number of damaged video slices are significantly reduced. Dual queues with cyclic service at the multiplexer also reduce the delays experienced by RT cells. The QoS for all categories of traffic is preserved.