Symmetric Peer-to-Peer Applications

In real-time networked applications, such as shared documents, users can interact remotely and yet share the exact same experience as if they were in a single machine. In this work, we propose a middleware for symmetric peer-to-peer applications in which decentralized instances can trigger asynchronous events and yet conform to identical behavior. Peers are allowed to join and leave the network at any time. Application developers must adhere to a restricted API, which is deterministic, stateless, and only supports pre-allocated memory. The middleware is responsible for synchronizing the events in a global shared timeline across the network. Also, based on memory snapshots and deterministic simulation, a "time machine" can rollback conflicting peers to resynchronize their state. We show that the middleware can handle applications with over 20 peers in a heterogeneous topology under high churn. For instance, in a simulation scenario of 25 events per minute with a network latency of 50ms, the peers need to roll back and resynchronize only 3% of the time.