ppOpen-HPC
ppOpen-HPC
http://ppopenhpc.cc.u-tokyo.ac.jp/
Open Source Infrastructure for Development and Execution of Large-Scale Scientific Applications on Post-Petascale Supercomputers with Automatic Tuning (AT)
Recently, high-end parallel computer systems have become
existing practical application code. The automatic tuning
larger and more complex. Yet, it is very difficult for scien-
(AT) capability is important and critical technology for
tists and engineers to develop efficient application code,
further development of new architectures and maintenance
that can take advantage of the potential for performance
of the framework.
“ppOpen-HPC” is a cooperative effort of the follow-
improvement of these systems.
We propose an open source infrastructure for devel-
ing six institutions, which include four departments of the
opment and execution of optimized and reliable simulation
University of Tokyo:
code on large-scale parallel computers. We have named

The University of Tokyo
this infrastructure “ppOpen-HPC”, where “pp”
User’s Program
stands for “post-peta”. The target system is the Post
T2K System based on many-core architectures (e.g.
Intel MIC/Xeon Phi), which will be installed in
ppOpen-APPL
FEM
FDMii
FVM
BEM
ppOpen-MATH
MG
ii
GRAPH
VIS
MP
ppOpen-AT
STATIC
ii
DYNAMIC
ppOpen-SYS
COMM
ii
FT
DEM
FY.2015. Its peak performance is expected to be
20-30 PFLOPS. “ppOpen-HPC” is part of a
five-year project (FY.2011-2015) spawned by the
“Development of System Software Technologies for
Post-Peta Scale High Performance Computing” effort funded by JST/CREST (Japan Science and
Technology Agency, Core Research for Evolutional
ppOpen-HPC
Science and Technology).
The infrastructure consists of various types of
libraries for scientific computation. Source code
Optimized Application with
Optimized ppOpen-APPL, ppOpen-MATH
developed on a PC with a single processor is linked
with these libraries, and the parallel code generated
is optimized for a post-peta scale system. The
framework covers various types of procedures for
scientific computation, such as parallel I/O of data-sets, matrix-formation, linear-solvers with practical and scalable preconditioners, visualization,
adaptive
mesh
refinement
and
dynamic
FEM
Finite Element Method
FDM
Finite Difference Method
load-balancing. This is useful in various types of
FVM
Finite Volume Method
computational models, such as FEM, FDM, FVM,
BEM, and DEM. This type of framework will provide dramatic improvements in efficiency, portability, and reliability in the development and execution of scientific applications. It reduces both the
number of steps in the source code and the time.
BEM
Boundary Element Method
DEM
Discrete Element Method
required for parallelization and optimization of
legacy code. Each component was developed based on
ppOpen-HPC http://ppopenhpc.cc.u-tokyo.ac.jp
 Information Technology Center (ITC)
A set of libraries related to node-to-node communication
 Atmospheric & Ocean Research Institute (AORI)
and fault-tolerance.
 Center for Integrated Disease Information Research
(CIDIR)
Libraries
 Graduate School of Frontier Sciences (GSFS)


in
ppOpen-APPL,
ppOpen-MATH,
and
ppOpen-SYS are called from user’s programs written in
Academic Center for Computing and Media Studies
Fortran and C/C++. All issues related to hybrid parallel
(ACCMS), Kyoto University
programming are “hidden” from users. ppOpen-AT also
Japan Agency for Marine-Earth Science and Tech-
optimizes load-balancing between the host-CPU and ac-
nology (JAMSTEC)
celerators/co-processors on heterogeneous computing
The expertise provided by researchers at the above institu-
nodes. Developed libraries of ppOpen-HPC will be open
tions covers a wide range of disciplines related to scientific
for public. Permission is granted to copy, distribute and/or
computing, such as system software, numerical li-
modify software and document of ppOpen-HPC under the
brary/algorithm research, computational mechanics and
terms of The MIT license. ppOpen-HPC will be installed to
earth sciences. Six key issues for algorithms and applica-
various types of supercomputers, and will utilized for re-
tions directed towards post-peta/exa scale computing are as
search and development using large-scale supercomputer
follows:
systems. Moreover, ppOpen-HPC will be introduced to

Hybrid/Heterogeneous Architectures
graduate and undergraduate classes of universities for de-

Mixed Precision Computation
velopment of parallel simulation codes for scientific

Auto-Tuning/Self-Adaptation
computing.

Fault Tolerance

Communication/Synchronization Reducing Al-
3Q 2012 (Ver.0.1.0)
gorithms


Reproducibility of Results
ppOpen-HPC for Multicore Clusters (K computer,
T2K, Cray etc.)
In ppOpen-HPC, these six issues are carefully considered.

ppOpen-HPC includes the following four components:
3Q 2013 (Ver.0.2.0: Current Release)
ppOpen-APPL

ppOpen-HPC for Multicore Clusters/Intel Xeon/Phi
A set of libraries corresponding to each of the five methods

New Features
Preliminary version of ppOpen-AT/STATIC

noted above (FEM, FDM, FVM, BEM, DEM), providing
I/O, optimized linear solvers, matrix assembling, AMR,
Both of Flat MPI and OpenMP/MPI Hybrid are
available in all components of ppOpen-APPL

etc.
ppOpen-MATH
Performance of ppOpen-MATH/VIS-FDM3D is
much improved.

A set of common numerical libraries, such as multigrid
ppOpen-MATH/MP-PP (tool for generation of
solvers, parallel graph libraries, parallel visualization, and
remapping table in ppOpen-MATH/MP) is
a library for coupled multi-physics simulations.
newly developed.

ppOpen-AT
AT enables smooth and easy shifts to further development
New functions (loop sprit, loop fusion, reordering statements) are added to ppOpen-AT
on new and future architectures, through use of
3Q 2014
ppOpen-AT/STATIC and ppOpen-AT/DYNAMIC. A spe-

cial directive-based AT language is developed for specific
3Q 2015
procedures in scientific computing, focused on optimum

Final version for Post-Peta-Scale System
memory access.

Further optimization on the target system
Prototype for Post-Peta-Scale System
ppOpen-SYS
Kengo Nakajima
Supercomputing Research Division
Information Technology Center
l
The University of Tokyo
Email: nakajima@cc.u-tokyo.ac.jp