Session VI: Cloud Native: Navigating the Traps for High Performance Computing Users

The emergence of cloud native technologies is changing the way applications are developed, deployed, debugged, and maintained. These changes come with various pitfalls particularly for high performance computing systems that traditionally provide their user base with the highest levels of compute, network, and storage performance. This talk will highlight new technologies in the cloud community with special focus on the benefits and challenges of deploying in an HPC data center. Tools such as Docker and Singularity, provide a framework for developers to build applications in isolation with the freedom to customize their development and runtime environments without impacting other users on the system. This can lead to a much more efficient application integration and deployment cycle characterized as continuous integration and continuous deployment (CICD) in cloud native terms. The broad acceptance of container orchestration services such as Kubernetes has encouraged the development of more sophisticated HPC workflows composed of simulations, AI, analytics, and other long running services that can be shared by multiple applications. But these enhancements in development and deployment features can come with a negative impact on overall application performance. How system features such as high-speed networks, high-speed filesystems, and low-jitter can be preserved in this environment is a key question that needs to be answered before cloud native becomes widely accepted in the HPC community.

Download Presentation

Location: Grand Ballroom C Date: March 29, 2019 Time: 8:40 am - 9:00 am Bruce D’Amora, IBM