I am interested in various scheduling and resource partitioning problems in distributed computing.
I am very interested in optimizing the execution of applications and understand the problems application users face when running their code on large-scale platforms.
I use a theoretical approach in which I design models and approximation algorithms to derive cost-efficient scheduling strategies for various types of applications.
My PhD focused on resource partitioning, design of scheduling strategies as so as data movement in HPC applications.
I currently work on two complementary research directions.
In the first direction, I am interested in modeling and studying in situ processing, where simulation and analytics are done in parallel to avoid I/O congestion.
This work was done in collaboration with Bruno Raffin, and our results have been published in IJHPCA journal.
The second direction concerns scheduling problems for stochastic applications, for instance Neuroscience applications.
The goal is to propose optimal scheduling strategies for single-core stochastic application running on a reservation-based platform.
Stochastic jobs are jobs whose execution time cannot be determined easily, mostly because they are input-dependent.
They arise from the heterogeneous, dynamic and data-intensive requirements of new emerging fields such as neurosciences
I spent two months in Vanderbilt University, Nashville, USA working on this topic in the group of Padma Raghavan.
Two papers on these topics have been published in IPDPS'19, IPDPS'20, and a journal paper in IEEE Transactions on Parallel and Distributed System.
I am currently working on related topics in the computing center IN2P3 Computing Center (Villeurbanne, France) of CNRS.
I am involved in the performance evaluation of in situ scheduling strategies on real-world applications and platforms. In this project, I developed a faithful simulation framework based on the SimGrid toolkit. This software is able to run the original code of real applications and allows to easily evaluate mapping and scheduling strategies for the different components of in situ workflows. A publication associated to this project is currently under review (see preprint).
Recently, I have been developing optimal algorithms for the scheduling of user requests on magnetic tapes. Such storage system is still widely used in computing center to save long-term data. Deciding the ordering of the user requests on the different tapes of the system is crucial to improve the quality of service experienced by users of tape storage systems. A paper on this topic is currently under review, with an associated preprint freely available online.
Do not hesitate to contact me if you want to collaborate or discuss on these topics!!