SCHEDULE: NOV 10-16, 2012
When viewing the Technical Program schedule, on the far righthand side is a column labeled "PLANNER." Use this planner to build your own schedule. Once you select an event and want to add it to your personal schedule, just click on the calendar icon of your choice (outlook calendar, ical calendar or google calendar) and that event will be stored there. As you select events in this manner, you will have your own schedule to guide you through the week.
Parallel Programming with Migratable Objects for Performance and Productivity
SESSION: Parallel Programming with Migratable Objects for Performance and Productivity
EVENT TYPE: Tutorials
TIME: 8:30AM - 12:00PM
Presenter(s):Laxmikant V. Kale, Eric J. Bohm
In this tutorial, we describe the migratable, message-driven objects (MMDO) execution model, which allows programmers to write high performance code productively. It empowers an adaptive runtime system (ARTS) to automate load balancing, tolerate faults, and support efficient composition of parallel modules. Multiple factors have led the HPC community to consider alternatives to SPMD programming: the immense scale of supercomputers, architectural heterogeneity, and the growing sophistication of scientific codes, with multiple modules and complex dynamic behaviors. With MMDO, an algorithm is over-decomposed into objects encapsulating work and data. Objects are message-driven and communicate asynchronously, to automatically overlap communication with computation. MMDO also allows the ARTS to manage program execution. For instance, the ARTS can migrate objects to balance load and optimize communication, resurrect objects on healthy processors to recover from hardware faults, and aggregate computationally intensive work to effectively utilize heterogeneous processing elements. Attendees will gain practical experience with the MMDO paradigm through a number of examples written in the CHARM++ programming system. CHARM++ supports several scalable applications, and has been deployed effectively on multicore desktops and 300K core supercomputers alike. Therefore, it provides a mature and robust vehicle for the exposition of MMDO design principles.
Laxmikant V. Kale - University of Illinois at Urbana-Champaign
Eric J. Bohm - University of Illinois at Urbana-Champaign