Efficiency comparisons between different model-based and measurement-based connection admission control under heavy traffic

Abstract

This paper discusses efficiency comparisons between different connection admission control (CAC) schemes with high arrival rate of connection requests. Through numerical studies, we have observed that the aggregate traffic converges to Gaussian 'slowly'. That is to say that when the number of homogeneous established connections of a traffic type is below a threshold (specific to that type), the aggregate traffic will be non-Gaussian. However, it approaches Gaussian as the number of connections increases over the threshold value. Hence, a CAC scheme based on a Gaussian model is not applicable to this traffic because the model considers the aggregate traffic to be Gaussian even when the total number of connections is below the threshold. This is also true if the threshold is higher than the maximum connections that a link can serve while maintaining the required quality of service. In light of this observation, we propose an enhanced version of the Gaussian model-based CAC scheme, which is also efficient if the traffic is non-Gaussian. Moreover, we introduce a measurement-based counterpart of that approach. Simulation results demonstrate that the enhanced CAC schemes perform well for two different types of traffic: video and data traffic traces.