Self-Managing Sensor-Based Middleware for Performance Monitoring and Data Integration in Grids

This paper describes a sensor-based middleware for performance monitoring and data integration in the Grid that is capable of self-management. The middleware unifies both system and application monitoring in a single system, storing various types of monitoring and performance data in decentralized storages, and providing a uniform interface to access that data. We have developed event-driven and demand-driven sensors to support rule-based monitoring and data integration. Grid service-based operations and TCP-based data delivery are exploited to balance tradeoffs between interoperability, flexibility and performance. Peer-to-peer features have been integrated into the middleware, enabling self-managing capabilities, supporting group-based and automatic data discovery, data query and subscription of performance and monitoring data.