Delay Optimization of Coded Computing With Hierarchical Task Partitioning

Coded computing can combat the straggling effect in distributed computing, especially at the wireless network edge. This letter focuses on the delay optimization of coded computing by jointly considering communication and computation imperfections. We adopt the adaptive privacy-preserving coded computing (APCC) strategy and model the task offloading process as an M/G/1 queueing system. The task division and partitioning problem is formulated as a non-convex mixed-integer non-linear programming (MINLP) with the objective of minimizing the task completion delay. Two near-optimal and low-complexity iterative update algorithms are proposed. Simulation results demonstrate substantial reductions in task completion delay over benchmarks.