Hermes: A scalable event-based middleware

摘要: Large-scale distributed systems require new middleware paradigms that do not suffer from the limitations of traditional request/reply middleware. These include tight coupling between components, a lack information filtering capabilities, and support for one-to-one communication semantics only. We argue event-based is scalable powerful type building large-scale systems. However, it important an platform includes all standard functionality application programmer expects In this thesis we describe design implementation Hermes, distributed, eventbased platform. The power flexibility Hermes illustrated throughout two domains: Internet-wide news distribution sensor-rich, active building. follows typeand attribute-based publish/subscribe model places particular emphasis on programming language integration by supporting type-checking event data inheritance. To handle dynamic, environments, uses peer-to-peer techniques autonomic management its overlay network brokers dissemination. Its routing algorithms, implemented top hash table, use rendezvous nodes to reduce state in system, fault-tolerance features repairing dissemination trees. All achieved without compromising scalability efficiency, as shown simulational evaluation routing. core extended with three higher-level services address different requirements computing environment. introduce novel congestion control service avoids broker during normal operation recovery after failure, therefore enables resource-efficient deployment expressiveness subscriptions enhanced composite performs detection complex patterns, thus taking burden away clients. Finally, security adds access according secure model. This supports fine-grained decisions so separate trust domains can share same network.