On Delay-optimal Issues Based on Queuing Approaches for Real-time Blending Services in Heterogeneous Access Networks

摘要

Heterogeneous network has become of particular interest recently, which has the potential to provide complementary features for overlapping coverage. Different from other works, we consider delay-optimal performance of heterogeneous networks and investigate it is average waiting time or average service time that contributes more for overall packet delay under different network conditions. A queuing model whose arrival process is a multi-state Markov-modulated Poisson Process (MMPP) and service process has two heterogeneous servers is established to depict heterogeneous access networks under Human to Human (H2H) and Machine to Machine (M2M) blending services. Our work starts by proposing a traffic model which is adopted to capture traffic behavior of M2M and H2H traffic over multiple time scales. Two queuing mechanisms aimed at minimizing delay, Shortest Average Waiting Time (SAWT) discipline and Shortest Average Service Time (SAST) discipline, arc presented. A cost-effective decomposition approach of infinitesimal generator matrix is proposed to equivalently divide the complicated generator matrix into a group of relatively simple submatrices in order to make the queuing model tractable. Numerical results show that the SAWT mechanism displays better performances with respect to robustness and stability. In terms of packet delay, SAST mechanism outperforms SAWT mechanism.