COLOC: Open workshop on data locality In conjunction with Euro-Par 2017, Santiago de Compostela, August 29, 2017

A well-known handicap for HPC applications running on modern highly parallelized and heterogeneous HPC platforms is that an increasing amount of time is spent in communication and data transfers; thus, it is necessary to design, implement and validate new approaches to optimize process placement and data locality management.

A well-known handicap for HPC applications running on modern highly parallelized and heterogeneous HPC platforms is that an increasing amount of time is spent in communication and data transfers; thus, it is necessary to design, implement and validate new approaches to optimize process placement and data locality management.

Program chairs

• Emmanuel Jeannot, Inria France
• François Verbeck, Atos/Bull, France

Topics

Areas or research interest include, but are not limited to:

• Modeling node topology
• Modeling network and communication
• Performance analysis of application to understand affinity
• Affinity metrics
• Runtime support for extracting affinity from application
• Code analysis in order to understand communication pattern
• Algorithm to improve locality
• Language, abstraction and compiler support for data locality
• Data structure and library support to better manage memory access
• Runtime-system and dynamic locality management
• System-scale locality optimization
• Validating locality optimization at thread or process level
• Memory management
• Locality management in large-scale application

Keywords: Algorithm, Application, Parallel Programming, Performance, and Scalability

Targeted audience:

• HPC application developers interested in exploring new ways to optimize their code.
• HPC centers and clusters managers to enhance cluster usage and application efficiency.
• Academics and researchers in scientific computing.

Program

• 9:00 - 9:15: Introduction. Emmanuel Jeannot and François Verbeck
• 9:15 - 10:00: Invited speaker - Managing data locality for convergence HPC/Big Data - Rosa Badia
• 10:00 - 10:30: Towards scalable parallel implementation of unstructured meshes - based Finite Element codes for Computational Fluid Dynamics. Quang Dinh, Nathalie Moller and William Jalby.
• 10:30 - 11:00: Break
• 11:00 - 11:30: Large-scale experiment for topology-aware resource management Yiannis Georgiou, Guillaume Mercier and Adèle Villiermet.
• 11:30 - 12:00: Runtime Support for Distributed Dynamic Locality. Tobias Fuchs and Karl Fürlinger.
• 12:00 - 12:30: Netloc: a Tool for Topology-Aware Process Mapping. Cyril Bordage and Brice Goglin.
• 12:30 - 13:00: Asynchronous One-Sided Communications for Scalable Fast Multipole Method in Electromagnetic Simulations. Nathalie Möller, Eric Petit, Quentin Carayol, Quang Dinh and William Jalby.

Important date:

• Workshop date: August 29, 2017
• Camera ready version: October 2017