Past event

The Society of HPC Professionals past event

Lunch & Learn – June 2020
Optimizing Seismic in the Cloud for Performance and Cost

Thursday, 25 June 2020
Live stream only due to COVID-19

Watch video and/or download presentation:

Download presentation 

Abstract:

In this talk I show how adoption of cloud enables novel optimization strategies for codes that improve not just time-to-solution, but also cost-to-solution.  In traditional HPC work, developers aggressively tune codes to run optimally according to the specifications of their on-premise computing platform, which was often procured in order to support a variety of computational workloads and may not be the best match for the application at hand.  When migrating HPC workloads “to the cloud,” the variety of resources available from providers offer a new opportunity to tune the selection of HPC hardware resources (such as processor, memory capacity, interconnect and storage) specifically for the workload.  I will use an example traditional seismic workload (RTM) to demonstrate how the flexibility of cloud HPC can enable unique optimization strategies that allow users to significantly reduce cloud costs, while maintaining performance requirements.  I will also discuss techniques on how to structure code for maximum flexibility with respect to these different cloud options.

Speaker Bio:

Branden Moore
HPC & Benchmarking Manager at NAG (Numerical Algorithms Group)

Branden Moore has over 15 years of HPC experience across academia, industry and government, focusing primarily on software optimization and architectural research. He joined NAG in 2018 to add his expertise to NAG’s international HPC consulting practice. Mr. Moore has worked with scientists across many different disciplines to enable them to solve real computational problems at speeds and scales previously out of reach. His background working at the intersection of software and hardware helps him to find the right software optimizations without negatively impacting the maintainability of real-world software projects. Having previously been at a large industrial research center and a U.S. national laboratory, Mr. Moore has an extensive history of working with legacy codes and languages, as well as developing new software. He has adapted codes to various accelerator platforms, such as GPUs and Cell Processors, as well as porting to different processors such as ARM and POWER, and supercomputers including IBM BlueGene and Cray XE/XT systems.