Computational Delay Improvement Techniques in UAV-Enabled MEC Networks

The emergence of complex software applications in recent years has significantly increased the demand for computational power. Hence, there is a need for a new computing platform that is easily accessible and can deliver a good Quality-of-Service (QoS) to its users. Mobile Edge Computing (MEC) is a promising solution, but efficient resource management algorithms are required to meet the QoS demands of the end-users. In this thesis, we propose two new Q-learning-based load-balancing resource management algorithms, QLLB and QLLB-PO, to reduce the computational delay in an MEC system. Unlike existing algorithms, the proposed algorithms use ideas from Active Monitoring Load-Balancing (AMLB) and K-th Nearest Neighbour (KNN) to optimally assign tasks to Unmanned Aerial Vehicles (UAVs), and Q-learning for the efficient CPU management at the MEC server level. We simulate and evaluate the proposed schemes in terms of performance and fairness against the baseline approaches. Our load-balancing algorithms, QLLB and QLLB-PO, show a considerable increase in the overall performance versus the baseline and Round-Robin based schemes. Under heavy load conditions, we observe as far as a 68% reduction in the total delay and a 50% reduction in the overall execution time with the QLLB algorithm versus the baseline algorithm. In addition, we observe as far as a 59% improvement in the combined UAV selection and the task data transmission delay. However, a drawback of the QLLB is that it drops tasks once its load-balancing component detects that the servers are fully loaded. With the QLLB-PO resource allocation algorithm, we are able to observe a slightly lower task drop rate versus the QLLB. In essence, we consider the task drop rate as a side-effect of load-balancing and a trade-off for the significant gain in performance that comes with QLLB and QLLB-PO. Based on the results, we state that our load-balancing algorithms perform well under load and are able to meet the QoS requirements of the end-users while keeping the servers in a healthy load-balanced state.