Alta Disponibilidade de um Gerenciador de Nuvem IaaS Baseada em Replicação: Experiência Resultados

• Gustavo B. Heimovski
• Rogério C. Turchetti
• Juliano A. Wickboldt
• Lisandro Z. Granville
• Elias P. Duarte Jr

Resumo

In this work we report the experience of adding a high availability solution to an IaaS (Infrastructure as a Service) cloud platform called Aurora. The proposed solution is based on the multi-master replication of the cloud manager, replicas form a cluster of cloud managers running on multiple datacenters. The implementation also includes a monitoring service for the replicas. The high availability solution is fully integrated to the cloud platform so that activation is done directly by pushing a button of the Graphical User Interface. The performance and robustness of the proposed solution were evaluated experimentally. The time to (1) incorporate a new manager instance to a cluster; (2) recover an instance after a failure and (3) replicate data in different scenarios were measured and are reported in this paper. The impact of the proposed solution cost is evaluated by measuring CPU and network usage. A stress test in which the link delay between two datacenters grows up to the operating limit of the replication solution is also reported. Finally, we present a table for the system availability as a function of the MTBF (Mean Time Between Failures).