Computational Intelligence Approaches for Energy-Aware Microservice Based SaaS Deployment in a Data Centre

Abstract

Microservice-based Software as a Service (SaaS) entails a software delivery approach in which software is constructed as a set of loosely coupled, independently deployable services known as microservices. Microservices are autonomous, small-scale services that collaborate to create a more extensive application. The microservice-based SaaS deployment problem refers to the challenge of efficiently deploying microservices within a SaaS to compute servers in a cloud data centre. This deployment leads to a significant increase in overall energy consumption, primarily due to the increased energy usage of the compute servers hosting the microservices. Moreover, the energy consumption of network devices that facilitate connections between in terconnected microservices also contributes to the overall energy usage. This increase in energy consumption raises concerns regarding sustainability, environmental impact and operational costs. In contrast to traditional SaaS deployment approaches, where energy considerations are frequently disregarded, this thesis addresses the energy increase associated with the deployment of microservice-based SaaS, focusing specifically on reducing the increase in energy consump tion in compute servers and network devices. To address this new microservice-based SaaS deployment problem, three Computational Intelligence (CI) approaches are designed and developed. First, an Adaptive Hybrid Genetic Algorithm (GA) is developed to tackle the problem. It achieves this by dynamically balancing the exploration-exploitation trade-off through an adaptive crossover rate. Furthermore, the adaptive hybrid GA incorporates a local optimiser, which refines the best solutions by improving the exploitation capacity of the adaptive hybrid GA. Second, a Hybrid Particle Swarm optimi sation (HPSO) approach is developed to address the new SaaS deployment problem. HPSO also integrates a local optimiser to enhance its exploitation capacity, thus further refining the best solutions. Additionally, HPSO has the capability to dynamically adjust the inertia weight and its cognitive and social parameters throughout the optimisation process. Third, an Ant Colony Optimisation (ACO) approach, equipped with new heuristic information, is developed to solve the new SaaS deployment problem. During the search for a compute server to host a microservice, this heuristic information aids the ants in selecting a compute server that will result in a lower increase in the energy consumption of both compute servers and network devices. In this thesis, a comparative study is conducted to evaluate the effectiveness, efficiency and scalability of the adaptive hybrid GA, HPSO and ACO approaches in solving the new SaaS deployment problem. The findings reveal that the adaptive hybrid GA is the most effective approach for minimising both total energy increase and the energy increase specifically in the compute servers. Its ability to provide energy-efficient solutions while maintaining good scala bility and fast execution times makes it the optimal choice for addressing the new microservice based SaaS deployment problem. The HPSO is identified as the second most effective and efficient approach, after the adap tive hybrid GA. It also demonstrates good scalability and faster execution times compared to ACO, efficiently generating optimal or near-optimal solutions as the problem size grows despite increasing complexity. Although the ACO is effective at minimising the increase in the energy consumption of network devices, it is the least effective approach for reducing the overall increase in total energy consumption. The cubic or quadratic increases observed in ACO’s execution times highlight its poor performance in scaling effectively with larger problem instances. The ACO’s poor scalability renders it impractical for handling larger problem sizes.