Epidemic Computing: a registration-free best-effort distributed computing model for unmanaged device crowds

Abstract

Crowds carry a large amount of idle computing capacity, but no entity can enumerate the devices. Managed distributed computing mobilizes only the registered fraction of the crowd, and most capacity remains unused when few register. We propose Epidemic Computing (EC), in which the reach of proximity communication triggers participation: recruitment spreads from device to device, without registration. EC abandons tracking and completion guarantees; the task's output includes only the results delivered by the deadline. Event-driven simulations show that, when the demand exceeds the registered capacity, EC delivers more total work than an idealized managed scheme even if half the crowd registered, and that this advantage persists under node churn. EC returns about 90% of a task that would take several minutes on a single device within an 8-second limit. This manuscript has been submitted to a journal of the IEICE.