Année : M2

Contact : Georges Da Costa

Email :

Organisation : IRIT

Projet : DVFS-Aware detection of HPC applications phases

Description du projet :

Phase detection is already well known on HPC applications. A classical example would be an application starting by a network phase (obtaining a dataset), followed by a computing phase (processing the data), and finishing with an I/O phase (saving the data). Phase detection for such applications are based on system monitoring, including hardware performance counters. Existing monitoring tools can obtain these values and convert them to a sequence of phases by identifying paterns of stable resource usage over time.

This project will evaluate how behave these phases at different frequencies, using multiple time series acquired from running HPC applications. The goal is to evaluate frequency-independent patterns of constant resource usage over time along with their energy and duration for each frequency.

The project is in the context of the ENERGUMEN ( funded by ANR) project at IRIT ( in Toulouse. For more information contact Georges Da Costa ( and Jean-Marc Pierson ( of the SEPIA team. The main research topics of the SEPIA time are optimisation of datacenter (multi-objective scheduling).

Expected ability of the student

  • Time series analysis
  • Linux
  • Knowledge on distributed systems and HPC systems would be a plus