Big Simulation and Big Data
Workshop
at Indiana University | January 9, 2017
1:00 PM, University Club, IMU
CREST Research in Dynamic Adaptive Methods for Extreme Scale Computation

Professor, Department of Intelligent Systems Engineering
Director, Center for Research in Extreme Scale Technologies
School of Informatics and Computing
Indiana University
Abstract
The Indiana University Center for Research in Extreme Scale Technologies (IU CREST) has been established to advance the state of the art in practical exposure and exploitation of runtime information for significant improvement in efficiency and scalability of current and future HPC systems as well as delivering superior introspective control for improved user productivity and performance portability. The center piece of this applied research is the proof of concept runtime system software, HPX+, that incorporates key distributed constructs to enable dynamic adaptive resource management and task scheduling. Other similar runtime systems such as Charm++ and OCR have evolved as well within a similar space although driven by concerns related to programming models and application requirements as their principal concern. HPX+ (initially from HPX and HPX-5) is itself motivated by specific challenges including: 1) a performance model, SLOWER, 2) architecture extensions to improve runtime operation, 3) support of irregular time varying graph-based applications both in the numeric and data-oriented such as adaptive mesh refinement and data analytics, and 4) an embodiment of a postulated cross-cutting execution model, ParalleX to provide a framework of guiding principles for the design, synthesis, and interoperability of the components of the hardware and software system stack. How this work serves current advanced evolutionary paths leading the US through the Hundred Petaflops performance regime and into the early Exascale era is important to contribute to the continued flux of innovations essential for avoiding stagnation due to possible limitations of incrementalism while providing useful enhancements of near term national goals and precluding the introduction of disruptive artifacts to ongoing achievements. The HPX family of runtime software is working closely with the Department of Energy labs and programs. HPX is being employed within the NNSA PSAAP-2 program as an important part of the C-SWARM shock-wave materials project. It is a component of the PNNL led Executions Model 2 project to devise methods for evaluating alternative paradigms. It served as a cornerstone of the Sandia driven XPRESS project to integrate OS, Runtime, instrumentation, control policies, and programming interfaces with key applications and DOE large scale systems. It is exploring its relationship to the Sandia-Livermore DARMA AMT project. And it is used as a testbed for parallel debugging for the Oak Ridge managed Hobbes OS project. It has also served in multiple NSF sponsored projects. This presentation will summarize the IU CREST research vision and findings in addressing the challenges confronting future generations of HPC at the end of Moore’s Law and the opportunities presented by dynamic adaptive computation.
Bio
Professor, Department of Intelligent Systems Engineering
Director, Center for Research in Extreme Scale Technologies
School of Informatics and Computing