Using Layered Bottlenecks for Virtual Machine Provisioning in the Clouds

Meeting the QoS objectives of fluctuating web workload requires techniques built on performance models, controller algorithms, monitors, etc. To meet the demands, we propose a controller algorithm using performance models that addresses the dynamic provisioning problem of multi-tier web applications in the cloud computing domain through addition of resources. The proposed algorithm aims to attain response time objectives by identifying "layered bottlenecks" and on this basis adding virtual machines (VM) and virtual CPUs, while keeping a check on limits such as spare VMs, processors-per-VM and replicas-per-VM. Here, Layered Queueing Network (LQN) performance models are used, alongside jLQNInterface, a tool developed in Java that allows solving, analyzing, and manipulating LQN models through the implemented API. The algorithm has been implemented using the tool and its applicability is demonstrated through a case study. By comparing two cases, it is shown that the proposed algorithm by using layered bottlenecks results in a model that satisfies the objectives with fewer resources.