On the elastic optimisation of cloud IaaS environments

• Main Author: Chatziprimou, Kleopatra
• Other Authors: Zschaler, Steffen ; Lano, Kevin Charles
• Published: King's College London (University of London) 2016
• Subjects: 004.67
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.689191

Elasticity refers to the auto-scaling ability of clouds towards optimally matching their resources to actual demand conditions. An important problem facing the infrastructure and service providers is how to optimise their resource configurations online, to elastically serve time-varying demands. Most scaling methodologies provide resource reconfiguration decisions to maintain quality properties under environment changes. However, issues related to the timeliness of such reconfiguration decisions are often neglected. A trade-o between the optimality of the reconfiguration solutions and the time cost to obtain these solutions is evident in the current literature. Highly accurate algorithms require a lot of data and time to execute, while more simplistic models may be fast to converge but provide poor quality solutions. In this thesis, we present a methodology for online optimisation of cloud configurations. Our motive is to balance the optimality versus timeliness trade-o in dynamic configurations management. We first employ a search-based approach to extract near-optimal configurations considering mutually conflicting performance and business quality attributes. Towards reducing the burden of time-consuming fitness evaluations of the configurations' quality during search-based optimisation, we develop surrogate models to predict the configurations' quality based on history observations. We evaluate our technique using CloudSim-based cloud simulation. Our experimental results show that the proposed methodology can produce high quality configurations with lead time of seconds and prediction error within 6%.