Round Robin Data Bases for Performance Evaluation of High Performance Applications and Cluster

TINETTI FERNANDO GUSTAVO; RIOS LEOPOLDO JOSE

Las Vegas

Conferencia; WorldComp16; 2016

We introduce a tool for automating (or aiding) performance evaluation of HPC (High Performance Computing) applications by combining both, Round Robin Databases (also referred to as RRD) and performance data collected at runtime. RRD monitoring is at the base of several well-known and popoular tools for cluster monitoring. We use take advantage of already existing RRD tools for collecting, processing, and presenting runtime data for scientific processing users. Scientific application (and even the hardware used by those scientific applications) are assumed to be performance optimized, but it is not always the case. Thus, collecting and analyzing runtime information will help scientific users to decide new optimizations and/or runtime strategies. The primary focus will be on parallel applications running in clusters, i.e. distributed hardware, since they are the most complex to optimize given their different and varying computing and communications patterns.