| Program |
- 12:00 - 12:50
Lunch in Nagoya University
- 13:00 - 13:05 Opening Talk
Prof. Takahiro Katagiri (Nagoya University)
- 13:05 -13:15 Dr. Fang-Pang Lin (National Center for High Performance Computing, Taiwan)
Title: NCHC overview
- 13:15 - 13:35 Dr. Fang-An Kuo (National Center for High Performance Computing, National Applied Research Labs, Taiwan)
Title: Large-Scale CFD Simulations with HPC Containers on NCHC HPC Systems
Abstract:
Presented is a novel computation framework for large-scale simulation of
general physical conservation laws in NCHC. The CFD framework code, named
UNICONES, along with pre-and post-processing tools provides high-fidelity
time-accurate flow simulations and allows user to load used-define plugins
to modify their physical conservation laws for space-time integration. The
code solver in UNICONES is based on the space-time conservation element and
solution element (CESE) method. It implements RANS models, such as SA, SST,
and K-epsilon models, and also implements a dynamic Smagorinsky
subgrid-scale model for large-eddy simulations (LES). The grid mesh system
is the unstructured mesh. To improve user experience on HPC systems, the
containerization of UNICONES imports more features including the
cross-platform execution and the automatic workflow through the job queuing
system built on HPC systems. A pre-compiled UNICONES code has been integral
in an HPC container of Singularity and can execute on multiple platforms
without re-compiling it.
- 13:35 - 13:45 Break
- 13:45 - 14: 05 Prof. Serge G. Petiton (University of Lille and CNRS, France)
Title: Challenges for Distributed and Parallel Very Sparse Matrix computing
Abstract:
Exascale machines are now available, based on several different arithmetic
(from 64-bit to 16-32 bit arithmetics, including mixed versions and some
that are no longer IEEE compliant) and using different architectures (with
network-on-chip processors and/or with accelerators). Brain-scale
applications, from machine learning and AI for example, manipulate huge
graphs that lead to very sparse non-symmetric linear algebra problems.
Moreover, those supercomputers have been designed primarily for
computational science, mainly numerical simulations, not for machine
learning and AI. New applications that are maturing after the convergence of
big data and HPC to machine learning and AI would probably generate
postexascale computing that will redefine some programming and application
development paradigms. End-users and scientists have to face a lot of
challenge associated to these evolutions and the increasing size of the
data.
In this talk, after a short description of some recent evolutions having
important impacts on our results, in particular about programming paradigms.
I present some results obtained on the still #1 supercomputer of the HPCG
list, Fugaku, for sequences of sparse matrix products, with respect to the
sparsity and the size of the matrices, on the one hand, and to the number of
process and nodes, on the other hand. Then, I introduce two opensource
generators of very large data, allowing to evaluate several methods using
very large graph-sparse matrices as data sets for several application
evaluations.
- 14:05 - 14:25 Prof. Nahid Emad (University of Paris Saclay/Versailles, France)
Title: A Parallel and Scalable Approach for High Performance Learning
Abstract:
This presentation highlights certain common characteristics and methods in
the field of high-performance numerical computing and that of machine and
deep learning. In this context, a new machine learning approach based on the
Unite and Conquer methods, used in linear algebra, will be presented. The
important characteristics of this intrinsically parallel and scalable
technique make it very well suited to multi-level and heterogeneous parallel
and/or distributed architectures. Experimental results demonstrating the
interest of these approaches for efficient data analysis in the case of
clustering, anomaly detection and road traffic simulation will be presented.
- 14:25 - 14:45 Prof. Masatoshi Kawai (Nagoya University, Japan)
Title：Dynamic Core Binding for Load Balancing of Applications Parallelized with MPI/OpenMP
Abstract
Load imbalance is a critical problem that degrades the performance of
parallelized applications in massively parallel processing. Although an
MPI/OpenMP implementation is widely used for parallelization, users must
maintain load balancing at the process level and thread (core) level for
effective parallelization. In this paper, we propose dynamic core binding
(DCB) to processes for reducing the computation time and energy consumption
of applications. Using the DCB approach, an unequal number of cores is bound
to each process, and load imbalance among processes is mitigated at the core
level. This approach is not only improving parallel performance but also
reducing power consumption by reducing the number of using cores without
increasing the computational time.
- 14:45-14:50 Closing Remarks
Prof. Takahiro Katagiri (Nagoya University, Japan)
|