Speaker: David McCaughan, HPC Analyst, SHARCNET, University of Guelph
Abstract:
In their seminal reference, Gropp, Lusk and Skjellum muse that “to pull a bigger wagon it is easier to add more oxen then to find (or build) a bigger ox”. Increasingly, the wagons for many researchers are computational problems and they are pulled by the computers on which we attempt to study or solve
them. Fortunately for all of us, computers just keep getting faster…right?
Parallel computing refers to the solution and/or simulation of problems by large numbers of processors (or computers) working together. Unfortunately, computer programs don’t always cooperate as automatically as the beasts that we use to pull a wagon, so it isn’t sufficient in most cases to simply add
computational resources to the pile and realize faster execution or increased resolution in our programs.
This talk is intended for those without any significant background in high performance computing and serves as an introductory survey of the essential concepts of computational parallelism. We will begin by motivating the need for parallelism, dismissing the notion that we can just wait for computers to
get faster and thus solve our biggest problems. We will then consider highly accessible examples of parallel problem solving, including live demonstrations on a portable supercomputer we’ve built for this purpose. Time permitting, we will conclude with a high level view of how parallelism can be applied to
existing problems in order to provide a level of literacy with the subject matter.