February 2017

ECP PI Meeting

Principal investigators for the US Department of Energy’s Exascale Computing Project (ECP) met on November 29 at Argonne National Laboratory (Lemont, IL) to make presentations and consider the ECP’s big picture.

ICL’s George Bosilca, Jack Dongarra, and Heike Jagode attended this gathering, which was limited to one representative per project.

Quad charts for each of the funded projects aided the identification of topics and individuals relative to potential collaborations, interdependencies, and networking. Following the presentation of the charts, multiple parallel group discussions during days two and three allowed the participants to explore ways in which their projects connect to other portions of the ECP.

PEEKS Kick-off Meeting

ICL’s Ichitaro Yamazaki and Hartwig Anzt traveled to Albuquerque, NM, for a visit with collaborators at Sandia National Laboratories (SNL) on December 5 to discuss milestones and the initial steps involved in the Production-ready, Exascale-enabled Krylov Solvers (PEEKS) project, which is part of the US Department of Energy’s Exascale Computing Project (ECP).

Main SNL collaborators Mike Heroux, Erik Boman, Mark Hoemmen, and Siva Rajamanickam, as well as other SNL personnel, also attended the meeting. Ichi and Hartwig’s presentations provided both introductory and update information concerning PEEKS.

As the trip drew to a close, the ICLers had Friday morning off, so they took the opportunity to visit some local attractions. Hartwig toured the National Museum of Nuclear Science & History, while Ichi rode the Sandia Peak tramway to the top of the Sandia Mountains.

TESSE Workgroup Meeting

Members of the Task-based Environment for Scientific Simulation at Extreme Scale (TESSE) project workgroup met in the Stony Brook University of New York (SUNY) Global Center in New York City on December 12 to discuss progress.

TESSE uses an application-driven design to create a general-purpose, production-quality software framework that attacks the twin challenges of programmer productivity and portable performance for advanced scientific applications on the massively parallel, hybrid, manycore systems existing today and those anticipated for the future.

George Bosilca, Thomas Herault, and Damien Genet represented ICL at the meeting. Attending from the SUNY Institute for Advanced Computational Science (IACS) were Robert J. Harrison, W. Scott Thornton, and Mohammed Mahdi Javanmard; and from Virginia Tech, Department of Chemistry, Edward F. Valeev. In addition, invited to speak on the first day were Barbara Chapman of SUNY IACS and Brookhaven National Laboratory, and Dmitry Lyakh of Oak Ridge National Laboratory (ORNL).

Barbara discussed OpenMP and programming models. Dmitry presented on his work at ORNL, the agnostic design for performance portability.

Also during day one of the meeting, George, Thomas, and Damien gave a presentation on the Parallel Runtime Scheduling and Execution Controller (PaRSEC) project. PaRSEC is a generic framework for architecture-aware scheduling and management of microtasks on distributed manycore heterogeneous architectures. The presentation focused on the programming model, interfaces, how to interact with the runtime, and the integration of what is known as the SUMMA operation.

Robert gave a talk on the overall TESSE project, and Edward discussed his code TiledArray, a modern C++ library that is scalable to tens of thousands of processor cores.

Open discussions were organized around two topics: the scientific application community and the interoperability of the programming model and runtime systems.

Day two involved a discussion of the low-level aspects of the project.

IPDPS 2017 Review

Lead reviewers, including ICL’s Piotr Luszczek, and the organization committee of the IEEE International Parallel and Distributed Processing Symposium (IPDPS) met on January 5 in Rosemont, IL, near Chicago, to select papers for the conference, which is set to take place May 29 to June 2, 2017, in Buena Vista Palace Hotel, Orlando, FL.

Three tiers of reviewers overall were involved in the preparatory activities this year for IPDPS, with the lead review tier making the final decisions. The lead review team was diverse, encompassing academia, industry, and the national laboratories. Piotr participated in the application and software tracks.

ECP CEED Meeting

On January 10 through 12, the kick-off meeting of the Exascale Computing Project (ECP) Center for Efficient Exascale Discretizations (CEED) convened at Argonne National Laboratory to familiarize the participants from CEED’s collaborating institutions with relevant software and technologies.

The ECP CEED project will develop the next-generation discretization software and algorithms that will enable a wide range of US Department of Energy and National Nuclear Security Administration applications to run on future hardware.

Twenty-eight people from all the institutions involved in CEED attended the kick-off meeting. The CEED team encompasses Lawrence Livermore National Laboratory, which is leading the project, and Argonne National Laboratory, five universities, and outside contributors. Azzam Haidar and Stan Tomov represented ICL.

Stan described the meeting as a very useful jumpstart for the project—from getting to know the variety of software products and technologies already developed, to learning more about the applications that will be targeted during the first year.

Among the already-developed software and technologies were PUMI, Holmes, Nek, MFEM, MAGMA, OCCA, PETSc, Hypre, and Spack. Receiving attention the first year will be ExaSMR and MARBL.

Stan said the meeting also afforded the opportunity to synchronize and start the work on initial milestones, collaborations, and interactions.

As a collaborator in CEED, ICL will be instrumental in identifying, developing, and optimizing tensor contractions that are essential building blocks for the applications. We will also take an active role in co-designing APIs with the LLNL scientists, external partners on the team, and vendors, and deliver a high-performance tensor contractions package through our MAGMA library.

At the meeting, Stan and Azzam presented the MAGMA library and its batched linear algebra capabilities. ICL will build on the MAGMA batched capabilities and be instrumental in developing a MAGMA tensor contractions engine to power main computational components of the high-order methods targeted in CEED. In addition, we will lead the hardware thrust in CEED, which will entail handling interactions with the exascale hardware vendors.

The types of applications that the CEED project will enhance include global climate modeling, turbulent combustion in internal combustion engines, nuclear reactor modeling, additive manufacturing, subsurface flow, and national security.

MPHC Program Meeting

Trieste, a beautiful city on the Adriatic coast of Italy, was the location January 10–13 for the Master in High Performance Computing (MPHC) program meeting, which involved extensive teaching and labs with hands-on exercises for six hours each day.

The activities covered computer architecture and hardware for HPC and upcoming exascale systems. ICL’s Piotr Luszczek taught asynchronous CUDA programming and gave a full rundown of overview, design, and implementation details of all the linear algebra libraries we produce and release at ICL.

The MHPC program is for graduate students from the International School for Advanced Studies (Scuola Internazionale Superiore di Studi Avanzati) and the International Centre for Theoretical Physics. Program links and materials are available here.

Piotr notes that Trieste is only a stone’s throw away from Slovenia and the Republic of Croatia.

ISTC for Big Data Tech Meeting

On January 25 at the Intel Science and Technology Center (ISTC) for Big Data technical meeting in Boston, MA, Piotr Luszczek and Thomas Herault discussed the integration of the linear algebra algorithms developed at ICL into the SciDB computational database management system. SciDB allows the mining of insights from diverse data sources such as genomics, clinical, and financial markets.

The meeting also featured Michael Stonebraker of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT), Tim Mattson and Jeff Pankhurst of Intel, and Jason Kitchen and James McQuestion of the Paradigm4 company.

Inaugural ECP Annual Meeting

PausePlayPlayPrev|Next

The first “All Hands” meeting to launch the US Department of Energy’s Exascale Computing Project (ECP) occurred at the Knoxville Convention Center from January 31 to February 2. It brought together a significant portion of the ECP team—between 450 and 500 people. ICL sent fifteen people to represent its seven ECP subprojects.

The goal of the meeting was to establish a shared understanding of the ECP’s overarching vision, with particular attention to fostering the ECP community, making progress on subproject integration, and promoting an ethos of holistic codesign among the different ECP groups—science applications, software technologies, hardware technologies, and facility operations.

The ICL participants had three objectives in mind as they fanned out across the breakout sessions led by the ECP groups: determine the needs of different science applications for ICL software technology, ascertain how best to integrate ICL software into the ECP software ecosystem, and understand how ICL would need to coordinate with the different DOE laboratories in the testing and deployment of the software its ECP projects will produce.