Hierarchical Multi-Domain Architecture in Cloud Datacenters: Predicting Transient Downtime of Client Services Abstract: A key aspect of framing cloud datacenters Service Level Agreements is the derivation of downtime probability distribution function of a set of provisioned servers over a service

A key aspect of framing cloud datacenters Service Level Agreements is the derivation of downtime probability distribution function of a set of provisioned servers over a service window. This is further complicated by the hierarchy of power, network and server domains within any datacenter. This hierarchy adds significant complexity in the estimation of transient distributions. The process of failures and repairs in each level of the hierarchy is different and all of these processes need to be considered comprehensively. We extend prior research on cloud service availability modeling to precisely estimate transient probabilities under an instantaneous checkpointing strategy for a flexible failure distribution model. The estimation algorithms are computationally efficient in solving practical problems and yield rich information on transient system behaviors. The integrated hierarchical model is general and extensible to any datacenter topology comprising of the three domains, using various server failure and repair processes. Index Terms – Cloud computing, Switch, Power, Fault-tolerant Systems, Markov Chains