|
Presentation
The 18th "Scheduling for Large-Scale Systems" workshop will take place from Tuesday, July 8 to Thursday, July 10 (after lunch), in the "salon des diplomés" room of the École de Technologie Supérieure de Montréal. This will be the eighteenth edition of this workshop series after Aussois (2004), San Diego (2005), Aussois (2008), Knoxville (2009), Aussois (2010 and 2011), Pittsburgh (2012), Dagstuhl (2013), Lyon (2014), Dagsthul (2015), Nashville (2016), Knoxville (2017), Berkeley (2018), Bordeaux (2019), Fréjus (2022), Knoxville (2023) and Aussois (2024).
Taking advantage of the vibrant ecosystem around Machine Learning in Montreal and the joint CNRS-McGill-ETS ILLS laboratory which hosts and supports the workshop, this year edition will be focused around "Scheduling and IA": we particularly welcome presentations that focus on scheduling problems that arise in machine learning, or conversely on how IA can help schedulers, but will also include presentations on classical subjects that have made the reputation of the workshop, such as resilience, storage and energy efficiency for distributed computing systems.
As in previous editions, the workshop will be structured as a set of thematic half-day sessions. Time will be dedicated to informal discussions and exchanges, and the participants are strongly encouraged to break up in smaller groups based on common research interests.
Attendance of the workshop is by invitation only, and there will be no registration fees. Lunch and interlude refreshments will be provided by the ILLS laboratory, as well as a welcome aperitif party on Tuesday evening and a social dinner on Wednesday. Participants will be responsible for covering their travel, lodging and dinners.
Preliminary Program
Tuesday, July 8.
- 9h-9h30: Workshop introduction
- 9h30-10h: Oliver Sinnen – HIOCS: Heuristic Inter-Operator Co-Scheduling Method for Efficient DNN Inference on GPUs
- 10h-10h30: Oana Balmau – Systems for ML (Scheduling, Storage, and Data pre-processing), slides
- 11h-11h30: Laercio Lima Pilla – Exploring scheduling solutions for Federated Learning training, slides
- 11h30-12h: Rafael Pinot - Federated Learning with Adversarial Nodes
- 14h-14h30: Ana Gainaru – Scheduling in-situ analysis tasks attached to high fidelity simulations in HPC, slides
- 14h30-15h: Romain Perreira – Tasking runtime system for GPUs; energy efficiency on Aurora
- 15h-15h30: Yves Robert – Partial Detectors Versus Replication To Cope With Silent Errors, slides
- 16h-16h30: Damien Lesens – Some theoretical results on SVD methods for KV cache compression, slides
- 16h30-17h: Julien Herrmann - Interpretability of LLM-evolved heuristics, slides
Wednesday, July 9.
- 9h-9h30: Maxime Darrin – Leveraging Expert Usage to Speed up LLM Inference with Expert Parallelism
- 9h30-10h: Julia Gusak – Optimizing neural networks training using different types of parallelisms (data/tensor/model/pipeline) and re-materialization
- 10h-10h30: Olivier Beaumont – Optimized Forward-Backward Rematerialization for Memory-Efficient Pipeline Parallel Training
- 11h-11h30: Jiaxuan Chen – Inference and Fine-Tuning Co-serving for LoRA-Adapted LLMs, slides
- 11h30-12h: Félix Wirth – Towards Parallel Transformer-Based Large Language Models for Fast Inference, slides
- 14h-14h30: Frédéric Vivien – Green Scheduling on the Edge, slides
- 14h30-15h: Anne Benoit – Carbon-Aware Workflow Scheduling with Fixed Mapping and Deadline Constraint, slides
- 15h-15h30: Rajini Wijayawardana – Eliminating Job Terminations in Variable Capacity Cloud Datacenters
- 16h-17h: Student talks
How to get there ?
- option 1: walk together from the ÉTS (approx. 45 minutes, departure around 17h30 in front of the conference room)
- option 2: bus 36 west (Monk), in front of the ÉTS to Notre-Dame / Turgeon (approx. 15 minutes, bus every 20 minutes)
Thursday, July 10.
- 9h-9h30: Florina Ciorba – Performance, Portability, and Sustainability for Large Scale Simulations, slides
- 9h30-10h: Taylan Özden – EquilibrIO: Taming the I/O Tides in High-Performance Computing, slides
- 10h-10h30: Camille Coti – Modeling the energy consumption of shared GPUs
- 11h-11h30: Lucas Perotin – A New Algorithm for Online Scheduling of Rigid Task Graphs with Near-Optimal Competitive Ratio, slides
- 11h30-12h: Maxime Gonthier – Deadline-Aware Scheduling of Mixed-Criticality Tasks, slides
Workshop ends after lunch
Workshop location
The workshop will take place on the campus of the ETS, in the Griffintown neighborhood. More precisely, it will be held in the "Salon des Diplomés" room, on the second floor of the "Maison des étudiants" building (1220 rue Notre Dame O., Montréal, QC H3C 1K5). When you enter this building, take the stairs in front of you (or the elevator on your right) to reach the second floor, the conference room will be on your right.
Participants can find hotel rooms or rooms via Airbnb in the neighborhood. The campus is also easily accessible to the subway system (station "Bonaventure" on the orange line) and participants may also choose to stay in the "Plateau" neighborhood (close to the "Mont Royal" station of the orange line), with plenty of restaurants and bars.
List of confirmed participants
- Oana Balmau (McGill University)
- Olivier Beaumont (Inria)
- Anne Benoit (ENS Lyon)
- Kessia Cavalcanti-Nepomuceno (ÉTS Montréal)
- Jiaxuan Chen (McGill University)
- Jacob Chmura (McGill University)
- Florina Ciorba (Univ. Basel)
- Camille Coti (ÉTS Montréal)
- Maxime Darrin (Mistral)
- Pierre-Louis Filoche (ENS Saclay)
- Ana Gainaru (Oak Ridge Nat. Lab.)
- Maxime Gonthier (University of Chicago)
- Yulia Gusak (Inria)
- Valérie Hayot-Sasson (University of Chicago)
- Julien Herrmann (CNRS)
- Neeraj Kumar (Université de Montréal, MILA)
- Damien Lesens (ENS Lyon)
- Laercio Lima-Pilla (CNRS)
- Loris Marchal (CNRS)
- Taylan Özden (TU Darmstadt)
- Romain Pereira (Argone National Laboratory)
- Lucas Perotin (Vanderbilt University)
- Pablo Piantanida (CNRS)
- Rafael Pinot (Sorbonne University)
- Yves Robert (ENS Lyon)
- Sara Rouhani (ÉTS Montréal)
- Oliver Sinnen (University of Auckland)
- Frédéric Vivien (Inria)
- Rajini Wijayawardana (University of Chicago)
- Félix Wirth (Kog)
- Suyuchen Wang (Université de Montréal, MILA)
- Xiaofeng Zhang (Université de Montréal, MILA)
Organizing committee:
Loris Marchal (CNRS, ILLS), Pablo Piantanida (CNRS, ILLS) and Yves Robert (ENS Lyon).
For more information, please contact Loris Marchal.
Acknowledgements:
This workshop is made possible thanks to:
- French National Center for Research (CNRS) through the International Research Laboratory ILLS
- École de Technologie Supérieure de Montréal
- Fonds de recherche du Québec