« Modeling Energy Consumption of Heterogeneous Numerical Infrastructures in HPC »

Energy consumption is one of the main limiting factor for the design and deployment of large scale numerical infrastructures [4]. The road towards "Sustainable Exascale" is a challenge with a target of 50 Gflops per watt. Energy efficiency must be taken into account and must be combined with other criteria like performance, resilience, quality of service. In particular, power consumption is critical and power peaks should be avoided to ensure the proper power supply of the large-scale infrastructure [2].

As platforms become more and more heterogeneous (co-processors, GPUs, low power processors...), a deep understanding on how application and services can efficiently run on such large scale infrastructures remains a challenge. Simulation of infrastructures (like by using the SimGrid framework [1]) makes a significant step towards this direction.

The post-doc will explore the energy consumption issues in large scale high performance heterogeneous computing platforms by measuring the actual energy consumption on real infrastructures (including Grid’5000 test-beds), and proposing end-to-end energy models for large-scale HPC infrastructures. These models will account for data movement and computations. Their integration within the SimGrid framework will allow to test what-if scenarios in order to determine for given HPC applications what would be the most suitable hardware architecture in terms of computation cores and network configuration.

Several performance and energy models will be explored depending on targeted computing resources mixing simulated GPUs and low power processors. The post-doc will manage, propose and analyze several experimental scenarios in order to validate proposed solutions and to design energy-efficient HPC applications.

The candidate will be located in Inria Myriads team (IRISA, Rennes, Anne-Cécile Orgerie as principal advisor) and co-advised with the Avalon team (LIP, Lyon with Laurent Lefevre co-advisor). This post-doc of 18 months will be part of the Inria International Project Lab HAC SPECIS (High-performance Application and Computers, Studying PErformance and Correctness In Simulation)[3].

References : [1] - SimGrid : http://simgrid.gforge.inria.fr/ [2] - Kun Tang, Devesh Tiwari, Saurabh Gupta, Ping Huang, Qi Lu, Christian Engelmann, and Xubin He, "Power-capping Aware Checkpointing: On the Interplay among Power-capping, Temperature, Reliability, Performance, and Energy", IEEE/IFIP International Conference on Dependable Systems and Networks (DSN), 2016. [3] - IPL HAC Specis : http://hacspecis.gforge.inria.fr/ [4] - Anne-Cécile Orgerie, Marcos Dias de Assunção and Laurent Lefèvre, "A Survey on Techniques for Improving the Energy Efficiency of Large Scale Distributed Systems", ACM Computing Surveys, Volume 46, Issue 4, December 2014

For applying to the position go to : https://jobs.inria.fr/public/classic/fr/offres/2019-01521 Contacts : Laurent Lefevre (laurent.lefevre@inria.fr) and Anne Cécile Orgerie (anne-cecile.orgerie@inria.fr)