Abstract
Recurrent workflows revisit mutable state after intervening computations. Retaining state at its owner permits compact update returns, whereas migration changes future access paths and memory occupancy. This paper studies joint execution and state residency for serialized workflows in computing networks. The formulation distinguishes intermediate data, full-state fetches, and differential updates, minimizing computation and communication cost under resource reservations, peak-memory limits, and a completion deadline. Reuse-Frontier Label Search (RFLS) represents each execution prefix by the last host, owners of states awaiting reuse, and accumulated resource consumption. Componentwise dominance and admissible completion bounds prune the search. The unrestricted algorithm is exact; a finite label budget yields a heuristic. On 24 default synthetic instances, RFLS reduces mean paired cost by 4.64% relative to greedy placement. With a label width of 64, it matches 58 of the 63 certified optima in the intermediate-size tests. Experiments extending to 60 invocations examine the effects of resource limits, state reuse, and search effort on placement cost and runtime.