LIMITLESS — LIght-weight MonItoring Tool for LargE Scale Systems

This work presents LIMITLESS, a HPC framework that provides new strategies for monitoring clusters. LIMITLESS is scalable light-weight monitor integrated with other runtimes in order to obtain holistic view of the system combines both platform and application monitoring. paper description novel components architecture, including approaches reaching higher scalability based on combination in-transit processing performance prediction. We also include methodology improving scheduling by means machine learning classifiers profiling. includes practical evaluation simulated real platforms, shows significant scalability, retrieving data capacity reduced overheads. Results show prediction techniques reduce communications number packets more than 90% average, fine-grain allows run applications shared nodes reducing makespan 25% saving resources.