Resource Management in Multi-Access Edge Computing (MEC)

This PhD thesis investigates the effective ways of managing the resources of a Multi-Access Edge Computing Platform (MEC) in 5th Generation Mobile Communication (5G) networks.
The main characteristics of MEC include distributed nature, proximity to users, and high availability. Based on these key features, solutions have been proposed for effective resource
management. In this research, two aspects of resource management in MEC have been addressed. They are the computational resource and the caching resource which corresponds to the services provided by the MEC.
MEC is a new 5G enabling technology proposed to reduce latency by bringing cloud computing capability closer to end-user Internet of Things (IoT) and mobile devices. MEC would support latency-critical user applications such as driverless cars and e-health. These applications will depend on resources and services provided by the MEC. However, MEC has
limited computational and storage resources compared to the cloud. Therefore, it is important to ensure a reliable MEC network communication during resource provisioning by eradicating the chances of deadlock. Deadlock may occur due to a huge number of devices contending for a limited amount of resources if adequate measures are not put in place. It is
crucial to eradicate deadlock while scheduling and provisioning resources on MEC to achieve a highly reliable and readily available system to support latency-critical applications. In this research, a deadlock avoidance resource provisioning algorithm has been proposed for industrial IoT devices using MEC platforms to ensure higher reliability of network interactions. The proposed scheme incorporates Banker’s resource-request algorithm using Software Defined Networking (SDN) to reduce communication overhead. Simulation and experimental results have shown that system deadlock can be prevented by applying the proposed algorithm which ultimately leads to a more reliable network interaction between mobile stations and MEC platforms.
Additionally, this research explores the use of MEC as a caching platform as it is proclaimed as a key technology for reducing service processing delays in 5G networks. Caching on MEC decreases service latency and improve data content access by allowing direct content delivery through the edge without fetching data from the remote server. Caching on MEC is also deemed as an effective approach that guarantees more reachability due to proximity to endusers. In this regard, a novel hybrid content caching algorithm has been proposed for MEC platforms to increase their caching efficiency. The proposed algorithm is a unification of a modified Belady’s algorithm and a distributed cooperative caching algorithm to improve data access while reducing latency. A polynomial fit algorithm with Lagrange interpolation is employed to predict future request references for Belady’s algorithm. Experimental results show that the proposed algorithm obtains 4% more cache hits due to its selective caching approach when compared with case study algorithms. Results also show that the use of a cooperative algorithm can improve the total cache hits up to 80%.
Furthermore, this thesis has also explored another predictive caching scheme to further improve caching efficiency. The motivation was to investigate another predictive caching approach as an improvement to the formal. A Predictive Collaborative Replacement (PCR) caching framework has been proposed as a result which consists of three schemes. Each of the schemes addresses a particular problem. The proactive predictive scheme has been proposed to address the problem of continuous change in cache popularity trends. The collaborative scheme addresses the problem of cache redundancy in the collaborative space. Finally, the replacement scheme is a solution to evict cold cache blocks and increase hit ratio. Simulation experiment has shown that the replacement scheme achieves 3% more cache hits than existing replacement algorithms such as Least Recently Used, Multi Queue and Frequency-based replacement. PCR algorithm has been tested using a real dataset (MovieLens20M dataset) and compared with an existing contemporary predictive algorithm. Results show that PCR performs better with a 25% increase in hit ratio and a 10% CPU utilization overhead.