Presentation Information
[N-2-13]Epidemic Computing: A Decentralized Computational Network Formation Method for Expanding Computational Resource Reachability
〇Takashi Nishitsuji1, Takuya Asaka2 (1. Toho University, 2. Tokyo Metropolitan University)
Keywords:
Distributed computing,Best effort,P2P,Ad-hoc network
Compute-intensive services such as generative AI concentrate load on centralized infrastructure like data centers. Cooperative use of capable mobile devices offers a low-cost, large-scale resource, but devices are hard to monitor fully and may vary in performance or leave, hindering both network construction and completion guarantees. We propose Epidemic Computing (EC): nearby mobile devices recursively re-delegate tasks, forming a distributed computing network that grows epidemically. Each node delegates only the part it cannot finish by its deadline, applied recursively to expand the system autonomously. EC is best-effort, tolerating missing results; target tasks are independently divisible and improve with completion ratio (e.g., Monte Carlo). Prior work is mostly single-hop or centralized, and multi-stage delegation has been rejected for reliability; EC enables it by decoupling result collection from spreading and allowing partial results. By event-driven simulation, we evaluate computational-resource reachability (active nodes) and the in-deadline completed work ratio (CWR) versus task size. EC expands active nodes from 18 to 77 and keeps CWR far above the baselines, thus extending reachability while sustaining higher in-deadline completion as the workload grows.