Eucalyptus-based private clouds: availability modeling and comparison to the cost of a public cloud

High availability in cloud computing services is essential for maintaining customer confidence and avoiding revenue losses due to SLA violation penalties. Since the software and hardware components of cloud infrastructures may have limited reliability, the use of redundant components and multiple clusters may be required to achieve the expected level of dependability while also increasing the computational capacity. A drawback of such improvements is the respective impact on the capital and increase in acquisition and operational costs. This paper presents availability models for private cloud architectures based on Eucalyptus platform, and presents a comparison of costs between these architectures and similar infrastructure rented from a public cloud provider. Metrics for capacity-oriented availability and system steady-state availability are used to compare architectures with distinct numbers of clusters. A heterogeneous hierarchical modeling approach is employed to represent the systems considering both hardware and software failures. The results highlight that improvements on the availability are not significant when increasing the system to more than two clusters. The analysis also shows that the average available capacity is close to the maximum possible capacity in all architectures, and that it takes 18 months, in average, for these private cloud architectures to pay off the cost equivalent to the computational capacity rented from a public cloud.