Learning a multi-criteria decision model for cost and performance optimization in the Cloud
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Cloud computing has emerged as a transformative force in modern computing, offering
ubiquitous access to on-demand services that cater to diverse client needs. As this paradigm continues
to permeate everyday life, the optimization of cloud provider characteristics has become paramount.
Combinatorial optimization, drawing upon discrete mathematics and computer science techniques,
serves as a vital mechanism to address real-life challenges in this domain. However, the proliferation
of numerous cloud providers, coupled with a myriad of selection criteria, has rendered decisionmaking
increasingly complex.With the growing array of cloud providers, it becomes increasingly
important to optimize the performance of these services to effectively meet the demands of users and
enterprises alike. This thesis presents novel methodologies aimed at enhancing cost-efficiency and
performance within cloud computing environments, harnessing sophisticated multicriteria decision
support systems. Indeed, we introduce two distinct contributions to address the challenges associated
with performance optimization.
The first contribution of this thesis addresses the pressing need for robust decision-making
frameworks in cloud service selection. By proposing a multicriteria decision support system model,
we introduce a novel approach that integrates a modified neural network architecture with two hidden
layers. In this model, the first layer employs the ELECTRE technique to assess cloud services based
on multiple criteria, while the second layer utilizes a modified Tabu search algorithm, leveraging a
weighted sum approach for efficient decision-making. Notably, the input for our modified neural
network is cloud performance, allowing stakeholders to navigate the complex trade-offs between cost
and performance effectively. Through this contribution, we aim to provide decision-makers with a
comprehensive framework to optimize costs and performances in cloud computing, thus enhancing the
overall efficiency and effectiveness of cloud service selection processes.
Building upon this foundation, the second contribution of this thesis introduces a pioneering
approach termed Multi-Objective Optimization Approach for Cloud Services Finding (MOOA-CSF).
By harnessing supervised learning and multicriteria decision techniques, MOOA-CSF aims to
optimize both price and performance in cloud computing environments. Through the utilization of an
artificial neural network (ANN) to classify cloud services based on their features, coupled with the
application of the ELECTRE method to order these services, MOOA-CSF offers a systematic
framework to identify optimal cloud solutions tailored to the specific needs of clients and systems.
Furthermore, by incorporating a modified genetic algorithm to generate hybrid cloud services,
MOOA-CSF extends the boundaries of traditional optimization techniques, offering innovative
solutions to complex optimization problems in cloud computing. Through rigorous evaluation using
diverse scenarios, simulation results demonstrate the efficiency and effectiveness of our approach,
underscoring its potential to enhance cloud computing performance optimization in real-world settings
significantly.