Coded Distributed Computing with Wireless Shared Heterogeneous Workers

Distributed computing can parallelize large-scale computing tasks. However, due to the straggler effect, the performance of distributed systems may be diminished. Coded distributed computing mitigates the straggler effect by introducing redundant loads. In this research, we investigate a wireless network comprising multiple masters and heterogeneous workers, with each master allocated a distinct task. We divide computing tasks into two types: long-term computing tasks and short-term computing tasks. For long-term computing tasks, we propose the Static Worker Assignment algorithm (SWA), which balances the assignment of workers and minimizes the single iteration time of all tasks. Comparing SWA with two reference benchmarks reveals significant speed improvements of up to 54.03% and 51.14% over the respective benchmarks. For short-term tasks, we propose the Dynamic Worker Assignment algorithm (DWA), which reduces task reloading time without significantly increasing single iteration time. Compared to reloading by SWA, the DWA method may result in a 0.89% increase in single iteration time, while the task reloading delay decreases by 52.70%.