Modeling a Flexible Replication Framework for Space-Based Computing

Large-scale distributed systems often require complex interaction among dynamically joining and leaving participants. Compared to classical approaches coordinated by a central authority, peer-to-peer systems have been shown to provide a highly scalable and flexible architecture for such scenarios. Coordination middleware like tuple spaces can help to unburden developers from coping with the complexity of distributed coordination by offering simple abstractions for the decoupled interaction of autonomous peers. However, a fault-tolerant peer-to-peer system can only be built if replication mechanisms exist to persist data on several peers at once. To enrich space-based middleware with a flexible replication mechanism, we have designed a generic, plugin-based replication framework that supports easy adaptation via configurable replication schemes. The framework may act as a testbed to analyze the efficiency and reliability of different replication strategies. Its architecture is built via highly composable coordination patterns that internally interact via space containers. Using the generic framework, this paper shows how different variants of multi-master replication can be realized and how they can be adapted for various scenarios.