An engineering firm runs large-scale computational fluid dynamics simulations that are fault-tolerant and support application-level checkpointing every 10 minutes. Individual jobs run for 6 to 12 hours across hundreds of vCPUs. Management's primary business requirement is to minimize compute cost while still completing each simulation batch within a 48-hour window. The workload has no fixed daily baseline; it runs in unpredictable bursts a few times per month. Which compute strategy best meets these requirements?