Model-based Tools and Techniques for Real-Time System and Software Health Management

Autonomous computing infrastructure imposes stringent requirements for consistency, synchronization, and security across multiple nodes, and manual monitoring of large-scale systems is neither scalable nor reliable. RFDMon innovates by leveraging Data Distribution Services middleware to decouple publishers and subscribers, enabling flexible and extensible monitoring system architecture. The hierarchical organization with regional coordination provides both scalability for large systems and resilience through distributed fault management without central bottlenecks.

This paper introduces RFDMon, a distributed event-based monitoring framework for autonomous systems that achieves real-time fault diagnosis and recovery through scalable hierarchical monitoring architecture. The system monitors infrastructure resources including CPU utilization, memory usage, network bandwidth, and hardware health across 100 to 800 computing nodes. It organizes monitoring into regions with local managers and global membership managers to enable efficient information dissemination while maintaining low communication overhead and minimal latency.

The framework successfully monitors large heterogeneous computing clusters with minimal resource consumption, automatically detecting infrastructure faults with low latency and minimal delay. Experimental deployments in production environments demonstrate effective scalability to hundreds of nodes with automatic self-reconfiguration when failures occur. The system enables diagnostic monitoring of infrastructure and can identify specific failed components while automatically adapting to infrastructure changes.