Rebecca Hartman-Baker

NERSC is not the first organization to implement multi-factor authentication (MFA) for its users. We had seen multiple talks by other supercomputing facilities who had deployed MFA, but as we planned and deployed our MFA implementation, we found that nobody had talked about the more interesting and difficult challenges, which were largely social rather than technical. Our MFA deployment was a success, but, more importantly, much of what we learned could apply to any infrastructure change. Additionally, we developed the sshproxy service, a key piece of infrastructure technology that lessens user and staff burden and has made our MFA implementation more amenable to scientific workflows. We found great value in using robust open-source components where we could and developing tailored solutions where necessary.

In this contribution, we discuss our experiences organizing the Best Practices for HPC Software Developers (HPC-BP) webinar series, an effort for the dissemination of software development methodologies, tools and experiences to improve developer productivity and software sustainability. HPC-BP is an outreach component of the IDEAS Productivity Project and has been designed to support the IDEAS mission to work with scientific software development teams to enhance their productivity and the sustainability of their codes. The series, which was launched in 2016, has just presented its 22nd webinar. We summarize and distill our experiences with these webinars, including what we consider to be “best practices” in the execution of both individual webinars and a long-running series like HPC-BP. We also discuss future opportunities and challenges in continuing the series.

The newest NERSC supercomputer Cori is a Cray XC40 system consisting of 2,388 Intel Xeon Haswell nodes and 9,688 Intel Xeon‐Phi “Knights Landing” (KNL) nodes. Compared to the Xeon‐based clusters NERSC users are familiar with, optimal performance on Cori requires consideration of KNL mode settings; process, thread, and memory affinity; fine‐grain parallelization; vectorization; and use of the high‐bandwidth MCDRAM memory. This paper describes our efforts preparing NERSC users for KNL through the NERSC Exascale Science Application Program, Web documentation, and user training. We discuss how we configured the Cori system for usability and productivity, addressing programming concerns, batch system configurations, and default KNL cluster and memory modes. System usage data, job completion analysis, programming and running jobs issues, and a few successful user stories on KNL are presented.