Dynamically Reconfigurable Distributed Modular Monitoring System for Supercomputers (DiMMon)

A design for a new dynamically reconfigurable distributed modular monitoring system framework is proposed in this paper. The proposed design allows combining both monitoring tasks (supercomputer ‘health’ monitoring and performance monitoring) in one monitoring system. Our approach allows different parts of the monitoring system process only the data needed for the task assigned to these parts. This helps to process a lot of performance data and to get information about dynamic features of heavy parallel tasks. Another feature of our framework is the ability to calculate performance metrics on-the-fly, dynamically creating processing modules for every job or other objects of interest.