Genomics Moves Us into Super-Computing… Learn to Grow Your Code

The era of faster single core processors ended a decade ago. Since then, manufacturers have increased the number of cores while maintaining the same overall power consumption. In order to recover performance, it is necessary to use many more cores simultaneously. This shift means writing software that can accommodate increased levels of parallelism as systems evolve towards low-powered many-core architectures.  Supercomputing centers such as the Oak Ridge Leadership Computing Facility have decades of experience on programming many cores simultaneously.  Developers of scientific software running at supercomputing centers must write robust code that can be easily ported to architectures that are not usually available to consumers; it must scale from a handful to tens of thousands of compute cores and adapt to diverse architectures, different memory capacities, network types, and disparate file systems. The transition to a hybrid accelerated system such as Titan, has shown that while developers are able to exploit much of the theoretical computational power, but that data movement has increasingly become the bottleneck even for well written software. This session will show best practices for "futurizing your code" to grow and run anywhere from local many-core desktops to small clusters, to clouds, to supercomputing centers.  The current state-of-the-art relating to analysis and processing of very large datasets will also be explored. Lastly, information on obtaining allocations at the center will be shared.