This is a full day workshop to expand the basic concepts of threads, message passing, and to introduce GPU programming to present multiple strategies in order to solve problems. The strategies are all MPI-based, but with a nod to OpenACC, which only exists at IU on the Big Red II supercomputer (Karst supercomputer is used for training). Both a lecture and a hands-on lab are included.
Objectives
- Be comfortable using MPI to solve parallel programming problems on a supercomputer
- Use heuristics to decide between the three primary methods of parallelization for a particular problem
- Know the pros and cons of both synchronous and asynchronous parallel algorithms
Prerequisites
- Basic UNIX/LINUX skills.
- Recommended: S4ES: Unix the Basics
- Comfort in using Karst or similar supercomputer
- Karst supercomputer account (https://kb.iu.edu/d/bezu#account)
- Ability to write and run a "Hello World" program in MPI
Agenda
- Run a simple MPI program for review.
- Explore various parallelization strategies.
- Choose one of the strategies and implement it into a program(s) previously run.
The Supercomputing for Everyone Series (SC4ES) aims to bring more users into the realm of advanced computing, whether it be visualization, computation, analytics, storage, or any related discipline. Research Technologies can take you to the next level of computing.
Supercomputing for Everyone Series workshops and seminars are led by personnel from Research Technologies, a division of University Information Technology Services and a center in the Pervasive Technology Institute at Indiana University.