Close this window

Email Announcement Archive

[Users] NERSC Weekly Email, Week of August 26, 2019

Author: Rebecca Hartman-Baker <>
Date: 2019-08-26 07:25:18

# NERSC Weekly Email, Week of August 26, 2019 # ## Contents ## - [Summary of Upcoming Events and Key Dates](#dates) - [Emergency Maintenance on Cori Begins at 8 am (Pacific) TODAY](#emergencymaint) - [Labor Day Holiday Next Monday; No Consulting or Account Support](#laborday) - [New MATLAB and Mathematica on Cori](#matlabica) - [ERCAP Submissions Now Open!](#ercap) - [Sign up Now for UPC++ Workshop on November 1 at NERSC!](#upcpp) - [Videos from June New Users Training Now Available](#newuservideos) - [Improve Your Job Start Performance by Following these Tips!](#jobtips) - [Use sshproxy to Minimize Time Spent on MFA](#sshproxy) - [No New "NERSC User News" Podcast this Week](#nopodcast) - [Come Work for NERSC!](#careers) - [Upcoming Outages](#outages) ## Summary of Upcoming Events and Key Dates <a name="dates"/> ## August 2019 Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 *26**27-*28*-29--30* 31 26 Aug Emergency Cori Maintenance [1] 27-30 Aug ECP/NERSC OpenMP Hackathon [2] 28 Aug SpinUp 1st Session [3] September 2019 Su Mo Tu We Th Fr Sa 1 *2* 3 4 5 6 7 2 Sep Labor Day Holiday [4] 8 9 *10--11--12* 13 14 10-12 Sep Cori Maintenance [5] 15 16 17 18 19 *20* 21 20 Sep ERCAP Office Hrs [6] 22 23 24 25 26 *27* 28 27 Sep ERCAP Office Hrs [6] 29 30 October 2019 Su Mo Tu We Th Fr Sa 1 *2* 3 *4* 5 2,4 Oct ERCAP Office Hrs [6] 6 *7* 8 9 10 11 12 7 Oct ERCAP Office Hrs [6] 7 Oct ERCAP Proposals Due [7] 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 Notes: 1. **August 26, 2019**: [Emergency Cori Maintenance](#emergencymaint) 2. **August 27-30, 2019**: [OpenMP Hackathon](#hackathon) 3. **August 28, 2019**: August SpinUp Workshop first day 4. **September 2, 2019**: [Labor Day Holiday (No Consulting or Account Support)](#laborday) 5. **September 10-12, 2019**: Cori Monthly Maintenance 6. **September 20, 27, October 2, 4, 7, 2019**: ERCAP Office Hours 7. **October 7, 2019**: ERCAP Proposals due 8. All times are **Pacific Time zone** ### Other Significant Dates ### - **Nov 1, 2019**: [UPC++ Training](#upcpp) - **Nov 28-29, 2019**: Thanksgiving Holiday (No Consulting or Account Support) - **Dec 24, 2019-Jan 1, 2020**: Christmas/New Year Holiday (Limited Consulting or Account Support) ## Emergency Maintenance on Cori Begins at 8 am (Pacific) TODAY <a name="emergencymaint"/> ## NERSC's unique filesystem usage patterns have exposed issues in the way that the new operating system on Cori interfaces with the filesystems. The August 9 emergency maintenance patched a bug that was causing some issues with the filesystems, and after that problem was corrected, an additional issue that had been masked by the original issue was exposed. We now have a working patch from Cray addressing this second issue that will be installed today, starting at 8:00 am Pacific time. Beginning at 8:00 am, users will no longer have access to Cori. There is a reservation in place to prevent jobs from running, but any errant jobs will be killed at that time. All login and compute nodes will be removed from service for the duration of the maintenance. In addition, the DTN systems will be unavailable for the duration of the maintenance in order to perform an update to the filesystems running GPFS. While the system is down, we will also take the time to run a brief acceptance test for our new community filesystem (coming soon!). Altogether we predict that this will be a 12-hour maintenance, but we hope to return Cori to users as early as possible. Thank you for your patience as we eliminate these bugs in the system! We are all looking forward to a stable, performant Cori system. ## Labor Day Holiday Next Monday; No Consulting or Account Support <a name="laborday"/> ## Next Monday, September 2, consulting and account support will be unavailable due to the Berkeley Lab-observed Labor Day holiday. Regular consulting and account support will resume the next day. Operations staff are available for urgent queries via 1-800-66-NERSC, Option 1, at all times. ## New MATLAB and Mathematica on Cori <a name="matlabica"/> ## To resolve issues with the transition to the new operating system on Cori, we've installed new versions of MATLAB and Mathematica. In addition to improved compatibility, the new installations also make new functionality available. Recent updates to MATLAB have emphasized capabilities for deep learning and manipulation of large datasets. The latest version of Mathematica brings expanded functionality in mathematics, geographic visualization, audio and image processing, and machine learning. MATLAB users can now run release R2019a in addition to the default R2016a. If no problems arise, the R2019a version will become the default after the next Cori maintenance (September 10-12). Users of the MATLAB Runtime should note that the R2016a runtime module name has changed to MCR_R2016a, and that the runtime for R2019a can be invoked by loading the new MCR_R2019a module. As for Mathematica, versions 11.3.0 and 12.0.0 are now available, with 11.3.0 as default. Version 12.0.0 will become the default after the next Cori maintenance if no critical problems arise. ## ERCAP Submissions Now Open! <a name="ercap"/> ## Submissions for the 2020 ERCAP allocations process are now open. Requests will be accepted through Monday, October 7. During September and October, NERSC will hold ERCAP Office Hours via Zoom to assist users with submitting their requests. Please listen to this NERSC User News [podcast]( for more information about ERCAP. ## Sign up Now for UPC++ Workshop on November 1 at NERSC! <a name="upcpp"/> ## UPC++ is a C++11 library providing classes and functions that support Partitioned Global Address Space (PGAS) programming. UPC++ provides mechanisms for low-overhead one-sided communication, moving computation to data through remote-procedure calls, and expressing dependencies between asynchronous computations and data movement. It is particularly well-suited for implementing elaborate distributed data structures where communication is irregular or fine-grained. The UPC++ interfaces are designed to be composable and similar to those used in conventional C++. The UPC++ programmer can expect communication to run at close to hardware speeds. In this tutorial we will introduce basic concepts and advanced optimization techniques of UPC++. We will discuss the UPC++ memory and execution models and walk through implementing basic algorithms in UPC++. We will also look at irregular applications and how to take advantage of UPC++ features to optimize their performance. This event can be attended on-site at NERSC or remotely via Zoom. The remote connection information will be provided to the registrants closer to the event. [Registration]( is required for this event and space is limited so please register as soon as possible. Registration closes when the limit is reached or on October 18, 2019. ## Videos from June New Users Training Now Available <a name="newuservideos"/> ## Did you miss the last new users training? Are you interested in brushing up on the basics about NERSC? The training for new users presented on June 21 was recorded, and videos from the training are now available. To view the playlist on YouTube, please visit <>. ## Improve Your Job Start Performance by Following these Tips! <a name="jobtips"/> ## NERSC encourages users to follow best practices to get the most out of NERSC resources. The [best practices]( webpage has some useful tips for users of all kinds: small jobs and big jobs; singleton and multiple jobs. We frequently update this page with new best practices. Here is a highlight of a few newer tips. - **Run jobs from scratch, not home.** NERSC filesystems are optimized for different purposes. The scratch filesystem, for example, is optimized for performance but not for storage, while the home filesystem is just the opposite. Therefore we recommend running jobs from your scratch filesystem rather than your home directory. - **Improve efficiency by preparing the user environment before running.** When compute nodes are allocated for a batch job, all the commands in the script that load modules, set up runtime environment variables, and generally prepare for the "srun" line, run on the first of the compute nodes allocated to a job. These nodes are optimized for computing, not for running these types of commands, which means that they are slower than they would be if they were run on a login node. Fortunately, there is a way to rewrite your script with the Linux here document to run those commands on the login node before submitting the job (pre-prepared with those commands). See [the section with this tip]( for more details. ## Use sshproxy to Minimize Time Spent on MFA <a name="sshproxy"/> ## Do you find it tedious to type your password and one-time password every time you log in to Cori? Use the "sshproxy" tool developed by NERSC and use your credentials only once in a 24-hour period! Using sshproxy, you can generate secure ssh keys with limited-time validity that can be used for logging into NERSC resources via ssh. You can then ssh to NERSC machines without any additional authentication until the key expires. Setting up sshproxy requires only a few simple steps. Please see the ["Using MFA with SSH"]( section of the NERSC MFA webpage for all the details, but as a brief summary: - Download and install the sshproxy client appropriate for your computer. - On Mac or Linux, simply type " -u username", and when prompted, your password plus one-time password, to generate the ssh key. Then, you will log in automatically, by typing "ssh -i ~/.ssh/nersc". (You can also make some changes in your computer's dotfiles and ssh configuration files to require even less typing.) - On a Windows machine using the PuTTY ssh client, the process is similar; use a command prompt to create the ssh key, and then use pageant to load the key into PuTTY before launching PuTTY. NERSC consulting and account support staff are happy to help you get this set up; if you run into any issues please submit a ticket at <>. ## No New "NERSC User News" Podcast this Week <a name="nopodcast"/> ## During the (northern hemisphere) fall season, the "NERSC User News" podcast will move to a fortnightly schedule. Therefore, there will be no new episode of the "NERSC User News" podcast this week. We encourage you to instead enjoy some of our most recent episodes and greatest hits: - [Optimizing I/O in Applications]( Listen to an I/O optimization success story in this interview with NERSC Data and Analytics Services Group's Jialin Liu. - [ERCAP Allocation Requests]( Learn how to get compute and storage allocations on NERSC resources for next year in this interview with NERSC User Engagement Group's Clayton Bagwell. - [Roofline Model for Application Performance]( NERSC Application Performance Specialist Charlene Yang discusses the roofline model for application performance: what it is and how it works, how to use it to improve your application's performance, and future directions in roofline model research. - [Parallelware Trainer; Manuel Arenaz Interview]( Learn about the Appentra Parallelware Trainer tool: how it can help you learn to code with OpenMP and OpenACC, the features of the tool, and how to use it on Cori. - [Profiling Codes with Cray Performance Tools]( Learn about why you would want to profile your codes and the tools for profiling provided by Cray from Cray senior principal engineer and senior manager Heidi Poxon. - [Energy Efficiency and Environmental Consciousness at NERSC]( Learn about all the energy efficiency work going on at NERSC from building energy efficiency expert Norm Bourassa. - [Getting a Machine from Contract to Reality]( Listen to Systems Department Head Tina Declerck describe the complex process of going from a contract with a vendor to a supercomputer on the floor in production for users. - [A Day in the Control Room]( In this interview, NERSC Site Reliability Engineer Owen James talks about what it's like in the NERSC control room and how they ensure that the systems stay up for you. - [NERSC Year in Review and Looking Forward]( NERSC director Sudip Dosanjh discusses the highlights of 2018 at NERSC, as well as what NERSC is looking forward to in 2019 and beyond. - [NESAP Postdocs]( Learn from NESAP postdoc Laurie Stephey what it's like working as a postdoc in the NESAP program at NERSC. - [May Quarterly Maintenance]( The first ever NERSC podcast, featuring James Botts from the Computational Systems Group explaining why it takes so long to perform a maintenance on a supercomputer. The NERSC User News podcast, produced by the NERSC User Engagement Group, is available at <> and syndicated through iTunes, Google Play, Spotify, and more. Please give it a listen and let us know what you think, via a ticket at <>. ## Come Work for NERSC! <a name="careers"/> ## NERSC currently has several openings for postdocs, system administrators, and more! If you are looking for new opportunities, please consider the following openings: - [Computer Systems Engineer]( apply your knowledge of HPC systems and applications to helping users, developing training, and increasing the usability of systems for users. - [NESAP for Simulations Postdoctoral Fellow]( work in multidisciplinary teams to transition simulation codes to NERSC's new Perlmutter supercomputer and produce mission-relevant science that truly pushes the limits of high-end computing. - [NESAP for Data Postdoctoral Fellow]( work in multidisciplinary teams to transition data-analysis codes to NERSC's new Perlmutter supercomputer and produce mission-relevant science that truly pushes the limits of high-end computing. - [NESAP for Learning Postdoctoral Fellow]( work in multidisciplinary teams to develop and implement cutting-edge machine learning/deep learning solutions in codes that will run on NERSC's new Perlmutter supercomputer and produce mission-relevant science that truly pushes the limits of high-end computing. - [HPC Storage Systems Analyst]( Help architect, deploy, and manage NERSC's storage hierarchy (including Burst Buffer, Lustre, and Spectrum Scale filesystems, and HPSS archives). - [Site Reliability Engineer]( Provide a variety of engineering support services in a 24x7 environment to ensure that NERSC is accessible, reliable, secure, and available to our scientific users. - [High Performance Computing Security Developer]( Use your computer and programming skills to protect large-scale network and supercomputing systems in an open science environment. - [Application Performance Specialist]( Help prepare large-scale scientific codes for next-generation high performance computing (HPC) systems. (**Note:** We have received reports that the URLs for the jobs change without notice, so if you encounter a page indicating that a job is closed or not found, please check by navigating to <>, scrolling down to the 9th picture that says "All Jobs" and clicking on that. Then, under "Business," select "View More" and scroll down until you find the checkbox for "NE-NERSC" and select it.) We know that NERSC users can make great NERSC employees! We look forward to seeing your application. ## Upcoming Outages <a name="outages"/> ## Visit <> for latest status and outage information. - **Cori** - 08/26/19 8:00-20:00 PDT, Scheduled Maintenance - 09/10/19 7:00-09/12/19 19:00 PDT Scheduled Maintenance *Please note updated start time* The batch system and cscratch1 will not be available until 19:00 PDT on Thursday 12 Sept. The login nodes will be returned to the users (without cscratch1) by 17:00 PDT on Tuesday 10 Sept. The exvivo and cgpu systems will not be available for running user jobs until cscratch1 becomes available - 19:00 PDT on Thursday 12 Sept. - **DTN** - 08/26/19 8:00-10:00 PDT, Unavailable _______________________________________________ Users mailing list

Close this window