Cost-Aware Dynamic Virtual Machine Purchase Plan Orchestrator for Multi-tier Cloud Applications

Cloud deployment and multi-tier architecture are popular for web applications. Infrastructure as a Service (IaaS) offers users different Virtual Machine (VM) instances with various capacities and prices which can be used as servers in each tier of web applications. Because of heterogeneousness of workload in each tier, it poses a great challenge to choose suitable VM purchase plans for each tier as to adapt the varying workload and reduce the cost of VM usage. In this paper, we propose a new approach which conducts cost-aware dynamic VM purchase plans for multi-tier cloud applications. Our approach first uses reinforcement learning to generate several candidate plans to adapt the varying workload and then employs linear programming to find the optimal one which is the most money-saving among the candidates. The experiment results witness the proposed method gains better performance than the traditional threshold method in meeting the Service Level Agreement (SLA) and cutting down the cost.