Intent-Based Allocation of Cloud Computing Resources Using Q-Learning

In cloud computing, resource allocation is a critical operation that is usually performed in an infrastructure-aware manner. In practice, however, users often are not able to specify accurately their requirements, while an infrastructure’s specific characteristics are not always known. In our work, we assume that users provide their workload requirements in an infrastructure-agnostic manner, by describing their intentions regarding the way the workload should be served, e.g., with high capacity, with low cost etc. Towards this end, we propose the use of a Q-learning based Reinforcement Learning (RL) methodology that translates the users’ intentions to efficient resource allocations in a cloud infrastructure. The proposed mechanism is able to improve continuously the allocation of resources based on the user satisfaction and the infrastructure’s efficiency. Simulation results showcase the applicability of this approach, investigating various aspects of it.