Dynamic data processing, backup and abort strategy for cloud computing systems

Cloud computing systems are vulnerable to random shocks during operation, which may lead to performance degradation, system failure, and data loss. Data backup and mission abort are two key measures for mitigating failure risk and reducing data loss. However, existing backup and mission abort strategies, largely based on periodic schedules or static abort thresholds, are often insufficiently adaptive under deteriorating system conditions and stochastic mission progress. This paper develops a state-dependent dynamic framework that jointly optimizes sequential backup and abort decisions at discrete decision epochs. At each decision epoch, the system dynamically determines whether to remain in the current operating mode, switch between data processing and data backup, or abort the mission, based on the current state defined by the processed workload, the backed-up workload, the elapsed time, and the cumulative shock count. Within this framework, a probabilistic state-transition model is established, from which recursive expressions for the mission success probability and expected loss are derived. The optimal policy and the corresponding optimal decision interval are then obtained by backward dynamic programming. Numerical results show that the dynamic policy outperforms benchmark static policies in terms of expected loss and exhibits clear structural properties. Robustness analyses under several extended settings further confirm the effectiveness and practical applicability of the proposed framework.