Speaking at the 2016 SEG, Leo Reiter, (Nimbix CTO) described the problems of processing seismics in the cloud. Seismic imaging is no longer ‘embarrassingly parallel’ and today’s commodity cluster is not up to the task. Reiter describes a ‘conflict of architectures’ with the cloud’s low interconnect bandwidth, loosely coupled architecture and poor GPU support. On the other hand an HPC seismic machine needs high bandwidth interconnect and plenty of GPUs.
Container technology brings the promise of ‘bare metal’ performance, potentially without the overhead of an operating system. In practice though, Docker-style containers use a virtual machine and lose some of these benefits.
Enter Nimbix Jarvice, a containerized runtime platform for HPC workflows. Jarvice containers run directly on bare metal and offer native performance, access to accelerators and to an InfiniBand-speed interconnect. The 2016 ‘Push-to-compute’ Jarvice edition allow continuous integration of containerized HPC applications with support for Nvidia and Docker images and ‘end-to-end, GitHub thro’ Docker to Jarvice’ deployment. Recently IBM and Nimbix announced availability of Minsky-powered GPU options on the Nimbix cloud.
© Oil IT Journal - all rights reserved.