Engineer F/M: Tracing and energy-aware performance monitoring of HPC/AI workloads

Context
Within the framework of the European project
EU-LAC Supercomputing Network for AI
(Team Europe Initiative "EU-LAC Digital Alliance"), coordinated by the Barcelona Supercomputing Center (BSC), in collaboration with INESC TEC (Portugal), FZJ (Germany), CSC (Finland), IZUM (Slovenia) and CINECA (Italy).
This project aims at strengthening cooperation between Europe and Latin America and the Caribbean (LAC) in high-performance computing (HPC) and artificial intelligence. Inria leads the work package on joint EU-LAC use cases and, more specifically, Use Case 1: a unified methodology for tracing HPC/AI applications and monitoring their performance and energy consumption. This methodology provides the common basis for the other use cases of the project (e.g. large language models adapted to Latin American language varieties, or earthquake and tsunami simulation) and will be deployed in LAC HPC centres selected through an open call.
Occasional travel in Europe and Latin America (project meetings, workshops, visits to partner centres) is expected.
Assignment
The recruited person will be responsible for designing and implementing a methodology and an open-source toolkit to collect, store and process HPC/AI application and system traces, with a particular focus on energy consumption (CPU/GPU power draw, node-level energy counters), I/O behavior and resource utilization. This characterization will rely on existing tools wherever possible, such as profiling tools (Darshan, Tau), tracing tools (Recorder), energy measurement interfaces (RAPL, NVML) and the IOPS experiment orchestration framework developed in the team, complemented by the components needed for portable deployment across European and Latin American HPC infrastructures.
In addition to that, the person will be responsible for instrumenting the executions of the other use cases of the project and providing them with performance and energy dashboards (time-to-solution, resource utilization, I/O pressure, carbon footprint). They will work with the leads of the other use cases (BSC, CINECA) and with the system administrators of the participating HPC centres, in particular the Latin American centres selected through the open call, whom they will support in adopting the methodology.
Finally, the expected results are the toolkit and dashboards, a methodology guide (including GDPR-aligned data anonymization and governance), and a public repository of anonymized traces.
For more information on the proposed research subject:
Examples of similar work conducted in the same research team:

• https://hal.inria.fr/hal-03753813
• https://hal.inria.fr/hal-03808833/
• F. Boito, L. Teylo, M. Popov, L. Aimi, A. Bandet, et al. TOTO: Transparent I/O Tuning for HPC Applications. ICS 2026, Belfast, United Kingdom. https://hal.science/hal-05662894
• F. Boito, L. Teylo, M. Popov, T. Jolivel, F. Tessier, et al. A Deep Look Into the Temporal I/O Behavior of HPC Applications. IPDPS 2025, Milan, Italy. https://hal.science/hal-04887809
• IOPS (Integrated Orchestration for Parametric Studies), a benchmark orchestration framework developed in the team: https://gitlab.inria.fr/lgouveia/iops

Main activities
Main activities:
Studying the state of the art and existing tools for tracing, profiling and energy measurement on HPC systems;
Development of the pipeline for collecting, storing and processing traces (energy, performance, I/O, resource utilization) and its open-source release;
Running applications and benchmarks on HPC systems using scripts, treating and plotting results;
Deployment of performance and energy dashboards for the other use cases of the project;
Supporting Latin American HPC centres in adopting the methodology.
Additional activities:
Writing the methodology guide, reports and research papers (LaTeX);
Curating the public repository of anonymized traces (Zenodo or equivalent).
Skills
Technical skills and level required:
scripting (Bash, Python, etc.);
Unix: command line, ssh, etc.;
C/C++;
a plus (not mandatory): using HPC systems, slurm, etc.;
a plus (not mandatory): profiling and tracing tools, energy measurement (RAPL, NVML), time-series databases and dashboards (Prometheus, InfluxDB, Grafana);
experience in research, especially in HPC, would also be a plus.
Languages: fluent English (working language of the project); Spanish or Portuguese would be a plus.