IaaS Cloud Availability Planning using Models and Genetic Algorithms
Matheus Torquato, Lucas Torquato, Paulo Maciel, Marco Paulo Amorim Vieira · 2019
One of the main goals of cloud customers is to improve the availability levels of their applications. Thus, Cloud Providers usually offer Service Level Agreements (SLAs) to meet the availability requirements of the customers. However, setting up reasonable availability SLAs is a challenging task due to the cloud environment complexity. High availability is also a challenge for small private cloud environments, which nowadays have to provide a high availability platform for the hosted applications. In this paper, we propose an approach to support the design of Infrastructure-as-a-Service (IaaS) Cloud architectures aiming at the desired levels of system availability. Our fundamental architecture considers the main four components of virtualized environments: Front-End, Physical Machines, Virtual Machines and sStorage Area Network. We designed an availability model for IaaS architectures based on these components and used it as input for our genetic algorithm (RENATA). RENATA output suggests redundancy schemes to achieve classes of target availability, from managed environments (with 99% of availability) to ultra-availability environments (with 99.99999% of availability). Our results also include the Capacity Oriented Availability of each redundancy scheme. We also present a failure and repair injection experiment to support the verification of model correctness.