Abstract
Cloud-based systems suffer from an increased risk of individual server failures due to their scale. When failures happen, resource utilization and system reliability can be negatively affected. Hybrid cloud models allow utilization of local resources in private clouds with resources from public clouds as and when needed through cloudbursting. There is an urgent need to develop cloudbursting approaches that are cognisant of the reliability and fault tolerance of external cloud environments. Recovery oriented computing (ROC) is a new approach for building reliable services that places emphasis on recovery from failures rather than avoiding them completely since even the most dependable systems will eventually fail. All fault tolerant techniques aim to reduce time to recover (TTR). In this paper, we develop a ROC-based fault tolerant approach for managing resources in hybrid clouds by proposing failure models with associated feedback control supporting a local resource-aware resource provisioning algorithm. We present a recovery-oriented virtual infrastructure management system (RVIMS). Results show that RVIMS is more reliable than those of single cloud environments even though TTR in the single cloud environments are about 10% less than those of RVIMS.