Online Flexible Resource Allocation

Project is the Dissertation work of Mark Towers for the University of Southampton on Online Flexible Resource Allocation in Mobile Edge Computing. The research can be found in the final_report folder while the code can be found in src and tests folder.

Mobile Edge clouds enable computational tasks to be completed at the edge of the network, without relying on access to remote data centres. A key challenge in these settings is that servers have limited computational resources that often need to be allocated to many self-interested users. Existing resource allocation approaches usually assume that tasks have inelastic resource requirements (i.e., a fixed amount of computation, bandwidth and storage), which may result in inefficient resource use and even bottlenecks. In this project, an elastic resource requirement mechanism is expanded upon to an online setting, such that tasks arrive over time with the prices and resource allocation determined by agents trained using reinforcement learning.

Within the project, a reinforcement learning environment is developed using the OpenAI gym environment specification in src/env. Numerous reinforcement learning algorithms are also implemented using Tensorflow 2: DQN, Double DQN, Dueling DQN, DDPG, TD3 and Seq2seq DDPG.