Dynamic task offloading and resource allocation for ultra-reliable low-latency edge computing

Applications such as online video, augmented reality/virtual reality (AR/VR) and Internet of Things (IoT) involve real time computation of high-size and large amount of data. This data have impact on users' quality of experience and system performance while being computed on resource-limited devices. In this regard, mobile edge computing (MEC) systems allow computations and traffic control can be performed at the network edge. However, MEC systems developed with respect to average-based performance fail to meet ultra-reliability and low-latency requirements which are receiving significant attention in 5G communication. In this thesis; C. -F. Liu, M. Bennis, M. Debbah ve H. V. Poor proposed a MEC system architecture in which imposed statistical constraints on the queue lengths with using extreme value theory within the context of ultra-reliable low-latency communication (URLLC) and resource allocation methods for local computation and task offloading to minimize users' power consumption in the network are investigated. Furthermore, a user-server association policy is studied by taking into account wireless channel quality, servers' existing workloads and computation capabilities. Referenced MEC architecture defines two-scale timeline for user-server association and dynamic task offloading. In this context, user-server association is executed in long/slow timescale; while dynamic task offloading is executed in short/fast timescale. In this thesis, proposed methods within the reference work are investigated and simulations are performed. Simulation results corroborate the efficiency of proposed methods providing ultra-reliable task computation and low-latency communication performance.