At the 2008 workshop, of what would later become known as the Rice University Oil & Gas HPC Conference, there were three presentations on use of GPUs for computational purposes. Since then, numerous examples of GPU-based HPC have figuratively paralleled the growing success of the RiceU event. It was with a touch of nostalgia then, that we approached one of our poster submissions for the 10th anniversary event.
Whereas load sensing GPUs is arguably table stakes for making use of them in workload-managed HPC clusters, it’s really not enough. Univa addressed unmet requirements through the introduction of Resource Maps (RSMAPs) in Version 8.1.0 of Univa Grid Engine in 2012. Workload managers are all about efficiently and effectively managing resources, yet the RSMAP introduction acknowledged that a more-generalized abstraction for representing resources was required. Although our poster, “GPUs as Workload-Managed Resources: Challenges and Opportunities for Applications in the Oil and Gas Industry”, at this year’s RiceU event captures the specifics, suffice it to say that RSMAPs support a broad-and-deep array of use cases for GPU-based HPC – even when multiple GPUs are densely packed like sardines into the chassis of a single server! And though RSMAPs have addressed most needs, we’re still enhancing support for GPUs via RSMAPs in Univa Grid Engine.
When it comes to visibility, you’ll be pleased to hear that our monitoring and reporting product is also GPU-savvy. This means the custom CUDA load sensor employed by Univa Grid Engine provides monitoring data that can be manipulated into tables and charts via Univa Unisight. Tables and charts for GPU-based workloads are also available via Univa Unisight as reporting data has been acquired from Univa Grid Engine.
From HPC to Big Data Analytics workloads arising from Machine Learning, to extension into private and public clouds, to emerging support for containerized GPU applications, there’s so much more to discuss. And based on our past experience, we have absolutely no doubt that the 2017 RiceU event will result in additional challenges and opportunities for all of us to address collectively as we employ GPUs for computational purposes in the oil and gas industry.
If you’re attending the Rice University Oil & Gas HPC Conference, we look forward to seeing you there – you’ll find us taking in talks and posters, as well as in the exhibits area. If you can’t make it to the event, please check back with us here at Univa as we’ll soon share a pointer to our GPU-centric poster. Cheers!