Keynote Talk (Opening Gala)
An Overview of High-performance Computing Today and a Look to the Future
Jack Dongarra, University of Tennessee Distinguished Professor
Monday, Nov. 16, 7:30 p.m.
This talk unveils the latest Top500 list of supercomputers, examines how HPC has changed during the last decade, and considers future trends. Changes will continue to have a major effect on software.
Enabling Scientific Discovery at NICS
Greg Peterson, Director, National Institute for Computational Sciences
Tuesday, Nov. 17, 11 a.m.
The University of Tennessee’s National Institute for Computation Sciences (NICS) is the advanced computing center of the Joint Institute for Computational Sciences (JICS). JICS seeks to lead the region and nation in scientific discovery and technological advancement. NICS spearheads those efforts for the community by providing and maintaining advanced computing resources, making a broad spectrum of applications ready, and training application developers and users to effectively exploit these resources.
PaRSEC: Programming Constructs for Efficient Applications @Scale
George Bosilca, Research Director, Innovative Computing Laboratory, University of Tennessee
Tuesday, Nov. 17, 2:30 p.m.
PaRSEC is a cutting-edge solution for addressing parallelism at scale, and an architecture-aware, task-based runtime, extended with several domain-specific languages. It allows developers to express their algorithms as data dependencies between parameterized tasks. In addition to discussing PaRSEC, this talk will present tools that allow for performance gathering and autotuning of the resulting algorithms.
Modern Autotuning for Hardware Accelerators
Piotr Luszczek, Research Director, Innovative Computing Laboratory, University of Tennessee
Tuesday, Nov. 17, 3:30 p.m.
Autotune has become an essential tool for performance engineering teams that tackle the complexities of hardware accelerators. Modern GPUs and coprocessors do not present themselves to the users through standard programming methods and require extensive tuning to achieve performance that can reach satisfactory levels. Going beyond that (closer to the peak performance) requires much more exhaustive search and rarely can be performed manually. This talk discusses how the process can be automated and applied to kernels found in scientific and industrial applications.
HPC Monte Carlo Simulations of Photon Propagation in Turbid Media
Dwayne John, HPC Consultant, User Assistance Group, National Institute for Computational Sciences
Wednesday, Nov. 18, 11 a.m.
The propagation of light in turbid media is an active area of research with relevance to numerous fields: biomedical diagnostics and therapeutics, weather and atmospheric science, astronomy, underwater and subterranean exploration, for example. This talk addresses the implementation of Berry Phase tracking at each photon scattering event and parallelization of the process using OpenMP and then MPI.
Fault-tolerant MPI Applications with ULFM
Aurelien Bouteiller, Research Scientist, Innovative Computing Laboratory, University of Tennessee
Wednesday, Nov. 18, 2:30 p.m.
As the number of components in HPC systems increases, probabilistic amplification entails that failures are becoming a common event in the lifecycle of an application. User-level Failure Mitigation (ULFM) is a minimalist but expressive extension to the MPI specification that provides the basic functionalities to restore post-failure communication capabilities. In this presentation, we present the novel features of ULFM-MPI, and embrace a holistic view to consider how both mainstream and boutique fault management strategies can benefit.
PAPI-EX: PAPI for the Future
Philip Mucci, Research Consultant, Innovative Computing Laboratory, University of Tennessee
Wednesday, Nov. 18, 3:30 p.m.
This talk presents PAPI-EX, the next generation of the Performance Application Programming Interface. The discussion will cover the status of current PAPI development as well as the high- and low-level milestones guiding PAPI into the exascale era.
UCX: An Open Source Framework for HPC Network APIs and Beyond
Pavel Shamis, System and Middleware Software Researcher, Oak Ridge National Laboratory, U.S. Department of Energy
Wednesday, Nov. 18, 4:30 p.m.
Unified Communication X (UCX) is a set of network APIs and their implementation for high-throughput computing. UCX comes from the combined efforts of national laboratories, industry, and academia to design and implement a high-performing and highly scalable network stack for next-generation applications and systems. UCX design provides the ability to tailor its APIs and network functionality to suit a wide variety of application domains and hardware. Pavel says that he and his colleagues envision that these APIs will satisfy the networking needs of many programming models, such as the Message Passing Interface (MPI), OpenSHMEM, Partitioned Global Address Space (PGAS) languages, task-based paradigms, and I/O-bound applications. This talk presents the recent advancement in the development of the UCX effort.
Also Speaking at SC15 from UT
Reuben Budiardja, Computational Scientist, Scientific Computing Group, National Institute for Computational Sciences
Tuesday, Nov. 17
1:30 p.m.–3 p.m.
Birds of a Feather, Understanding User-level Activity on Today’s Supercomputers with XALT, 13B
5:15 p.m.–7 p.m.
Posters, User Environment Tracking and Problem Detection with XALT, Level 4 Lobby
Friday, Nov. 20
8:30 a.m.–12 p.m.
HUST2015: Second International Workshop on HPC User Support Tools—Talk Title: Community Use of XALT in Its First Year in Production