A Distributed Architecture for Monitoring Private Clouds

Abstract

Cloud monitoring has become a key tool for organizations to ensure that their critical processes are being effectively managed in a private cloud infrastructure. A cloud monitoring tool should include complex tasks such as information extraction, management and planning of alerts in order to keep the service up during failures. However, current available solutions for private clouds only support some of these tasks. In this paper we present a distributed architecture for monitoring private clouds that unifies a set of monitoring tasks in a single service. This service includes the following configurable tools: Collectors, Metasensors, Distributors and Visualizer. A collector is a software tool that gathers monitoring information from physical and/or virtual machines (PM and/or VM) by using a set of special monitoring tools dubbed Metasensors. Collectors include a neural network-based component to classify the type of current workload on the monitored resource. This classification allows system administrators to define thresholds and create alerts to prevent saturation scenarios. A distributor is a tool that assigns the resources to be monitored (PMs and VMs) to collectors and concentrates the information gathered and processed by collectors in order to obtain the current workload status information of the private cloud infrastructure. This information can be showed to system administrator by using our visualizer tool. Our distribution approach avoids data centralization and provides the service with load balancing and fault tolerance. We developed a prototype based on our architecture and conducted an experimental evaluation as a proof of concept. The evaluation revealed the feasibility of our architecture in terms of performance and precision.