Storage-compute separation architectures are highly dependent on caching, and insufficient cache warm-up during service restarts, scale-out, or sudden changes in access patterns can cause latency spikes in the short term.