第16回ASE研究会のお知らせ

日時：2013 年 12 月 4 日（水）13:25 ～ 17:30（懇親会：18:00 ～）
場所：東京大学情報基盤センター（浅野地区）４階遠隔会議室（地図）
主催：東京大学情報基盤センター　スーパーコンピューティング部門

懇親会参加の方は、末尾の参加票をお送りください。

16th Advanced Supercomputing Environment (ASE) Seminar

December 4th (Wednesday), 2013
13:25 - 17:30

Information Technology Center (Asano Section), The University of Tokyo
4F Telecommunication Meeting Room（Map）

13:25 -　Opening

13:30 - 14:20　Invited Talk（１）

Invited Speaker

Dr. Francois-Henry Rouet (Lawrence Berkeley National Laboratory, USA)

Title

A parallel multifrontal solver that exploits hierarchically semi-separable representations

Co-authors

Xiaoye S. Li (LBNL), Artem Napov (University of Brussels), Jianlin Xia (Purdue), Maarten V. de Hoop (Purdue)

Abstract

Low-rank approximation techniques are an increasingly popular way of speeding-up sparse matrix algorithms.
We focus on direct methods for the solution of sparse linear systems, and we show how to embed hierarchically semi-separable (HSS) representations into a multifrontal solver.
These techniques allow to decrease both the operation complexity and the memory footprint of the solver for many real-life applications.
We focus on a new parallel geometric code that aims at solving discretized Helmholtz equations on regular meshes.
We present experimental results on very large domains (200+ million unknowns on 3D problems) and up to 16,000 cores.
We also present some new research directions, such as the use of randomized sampling.

14:20 - 14:30　Break

14:30 - 15:20　Invited Talk（２）

Invited Speaker

Dr. Samuel Thibault (University of Bordeaux, France)

Title

StarPU: leveraging clusters of heterogeneous machines through dynamic task scheduling.

Abstract

Heterogeneous accelerator-based architectures are more and more seen in production HPC clusters, featuring multicore CPUs, GPU accelerators, and now even Xeon Phi accelerators, thus providing an unprecedented amount of processing power per node.
It has thus become one of the biggest challenges in HPC to deal with such machines which expose such a highly unbalanced computing power.
To fully tap into the potential of these heterogeneous machines, pure offloading approaches, that consist in running an application on regular processors while offloading part of the code on accelerators, are not sufficient.
This talk will present the latest advances in the StarPU project, which aims at providing portable optimized performance on clusters of heterogeneous multicore+accelerator machines to task-based applications.
The goal is to relieve the programmer from the technical aspects of data management and task scheduling, while applying theoretical task scheduling algorithms on actual application execution to improve performance.
It also provides performance feedback through task profiling and trace analysis.

15:20 - 15:30　Break

15:30 - 16:20　Invited Talk（３）

Invited Speaker

Dr. Osni Marques (Lawrence Berkeley National Laboratory, USA)

Title

Performance of UPC Implementations of Matrix Computations and CG

Abstract

Partitioned Global Address Space (PGAS) languages offer programmers a shared memory view that can increase productivity and allow locality exploitation to obtain good performance on large-scale distributed memory systems.
It is expected that the PGAS model, as provided by Co-array Fortran, Chapel, X10, XcalableMP and Unified Parallel C (UPC), will play an increasing role in upcoming systems as it circumvents memory limitations that may be found in MPI.
In this presentation we show results for a parallel numerical library for dense matrix computations using UPC, and for linear solvers based on Cholesky and LU factorizations that are built upon that library.
We also show results for four different implementation of CG in UPC, using 1D and 2D distributions of the (sparse) matrix involved in CG computations, and evaluate those implementations on two different platforms by means of a set of matrices that exhibit distinct sparse patterns.

16:20 - 16:30　Break

16:30 - 17:20　Invited Talk（４）

Invited Speaker

Dr. Yifeng Cui (San Diego Supercomputer Center, USA)

Title

Earthquake Ground Motion Modeling: Petascale and Beyond

Abstract

The future exascale systems will exhibit a substantially different balance among processor speed, system interconnect and memory bandwidth.
Scientific applications must be radically re-designed and re-implemented for new programming models, new algorithms, and system tools to ensure a continued growth in sustained performance and productivity.
In this presentation, I will introduce a finite difference earthquake forward code to illustrate the challenges and opportunities on large-scale heterogeneous systems.
This GPU-based application implements high throughput, memory locality, communication reduction and communication/computation overlap.
These performance tuning made possible to achieve 2.3 Pflop/s sustained performance on 16,384 OLCF Titan GPUs, while simulating realistic 0-10 Hz earthquake ground motions relevant to building engineering design.
The simulations used a mesh comprising 443-billion elements in a calculation that includes both small-scale fault geometry and media complexity. Ground motion synthetics were computed using dynamic rupture propagation along a rough fault embedded in a 3D velocity structure with small-scale heterogeneities described by a statistical model.

17：30 -　Closing

18：00 -　Reception

-------------------------------------------------------------------------------
懇親会参加票
（ 11 月 25 日（月）12時までに、katagiri＠cc.u-tokyo.ac.jpまで送付 ）

　日時：2013 年 12 月 4 日（水）18：00 ～ ＠根津駅周辺
　予算：5,000 円程度

　ASE研究会懇親会に参加します

　お名前：
　ご所属：
-------------------------------------------------------------------------------

• センターユーザに限定せず、研究会は一般公開とします。
• 参加費は無料で、基本的に事前登録は不要です。

また、今後の開催予定を確実に知りたい方は、メーリングリストへの登録をお願いします。登録依頼については、下記問い合わせ先までお願いします。

〒113-8658　東京都文京区弥生2－11－16
東京大学　情報基盤センター

ASE研究会幹事　准教授　片桐孝洋
E-mail：katagiri＠cc.u-tokyo.ac.jp
（”＠”を半角にしてからお送りください。）