Análise de Disponibilidade de Servidores Virtualizados com Cadeias de Markov

Abstract

The analysis of availability of virtualized servers is an important tool for managers in information technology and communication especially, when it comes to planning and design of datacenters to provide many services for general companies. If the use of virtualization enables a cost reduction, it can also make the system more susceptible to downtime. This work analyzes the availability of two environments, one with a virtualized server and the other environment with non-virtualized servers. The services offered are e-mail, Domain Name System (DNS), Web Server and File Server, a typical scenario in many companies. It is developed a case study using analytical modeling with Fault Tree and Markov Chains. The Fault Tree is used to model the servers and Markov Chains to model the behavior of each component of hardware and software. The non-virtualized environment is composed of four servers, each one providing specific services, while the virtualized consists of a single server with four virtual machines, each one providing a service. By analyzing the models developed, the results show that although the non-virtualized system has less downtime, because has less dependence between the services, the difference in this case is 0.06% annually, becomes irrelevant when compared to the benefits brought by virtualization in the companies.