Avaliação de desempenho de virtualizações OpenVZ e KVM em nuvens computacionais usando Hadoop MapReduce

Abstract

Cloud computing provides access to a set of resources such as virtual machines, storage and network as services. In this context, virtualization has been used as a platform for resource-intensive applications, like Hadoop. The motivations of this use are the following features: server consolidation, scalability and better resources usage. OpenVZ and KVM are very popular and widely used virtualization platforms with distinct virtualization modes: container-based and full-virtualization, respectively. In this work, a set of benchmarks were conducted to measure the performance of Hadoop clusters deployed on OpenNebula private clouds with KVM and OpenVZ as virtualization platforms. The results show that the cluster hosted in the cloud with OpenVZ performs better than KVM in most tests, reaching a lower CPU virtualization overhead, higher disk I/O reading and memory throughput. The cluster hosted in the cloud using KVM, in turn, present better rates only in specifics tests of disk writing. The performance of network virtualization is very close to native in both clusters.