NERSCPowering Scientific Discovery Since 1974

Featured Announcements

Edison Big-Job Discount and Low Charge Class Retirement

February 27, 2017 by Rebecca Hartman-Baker

The big job discount on Edison will be reduced from 40% to 20% this Wednesday, March 1, 2017, and be removed entirely when Cori goes into production, currently planned for July 1.
Also this Wednesday, NERSC will retire the "low" charge class, which granted users a 50% discount in exchange for low priority. When a repo is out of time, its jobs will automatically be placed in the low-priority scavenger QOS and will be able to run free of charge if resources become available.
We expect that both of these changes will shorten wait times on Edison. The big-job and low-priority discount have the indirect effect of manufacturing additional allocated time. If a user has an allocation of one million NERSC hours and uses it entirely for large job runs, then they have consumed 1.67 million NERSC hours worth of resources while being charged for only one million. Likewise, one million NERSC hours of low-priority jobs consumes two million NERSC hours worth of resources. This inflation of allocation creates a situation where there are effectively more NERSC hours available to spend than can be delivered, resulting in a very crowded queue.
We originally implemented the big job discount as an enticement, to encourage users to run large jobs. However, given the level of adoption (more than half of NERSC hours spent on Edison were large jobs last year), the incentive was a success and can now be retired.
When we first implemented the low-priority job discount, we did not have a means for users whose allocations were low to run. This was seen as a way to stretch those last few allocation hours. Now that we do have a means for users to run after their allocation is used up, we no longer need the low-priority job discount.
All users, and especially those running big jobs, are encouraged to apply to the KNL Early Access Program. Jobs on the Cori KNL nodes will run free of charge through the end of June. Access will not be denied due to poor application performance; the purpose of the application process is to make sure that users are not unpleasantly surprised when running on the KNL nodes.

 

Read the full post

Allocation Year 2016 ends at midnight on Monday, Jan 9, 2017

January 9, 2017 by Clayton Bagwell

Allocation Year 2016 (AY16) ends at 23:59:59 on Monday, January 9, 2017. AY17 runs from Tuesday, January 10, 2017 through Monday, January 8, 2018.

Read the full post

NERSC Exascale Science Postdoctoral Fellowships for Data

December 13, 2016 by Rollin Thomas

We invite data science enthusiasts to apply for a postdoctoral position in our NERSC Exascale Science Applications Program (NESAP) for Data. NESAP for Data focuses on the needs of data-intensive science applications that perform processing and analysis of data from experimental or observational sources (for example: telescopes, microscopes, genome sequencers, light sources, or particle physics detectors). In collaboration with science user communities, NESAP for Data postdocs will explore how to adapt these applications to perform data science and knowledge discovery at scale on energy-efficient supercomputers. Our postdocs are also encouraged to participate in career, science enrichment, and outreach activities at Berkeley Lab.

Read the full post

Upgrades to Facility Will Require Multi-Day Outage in Early 2017

December 5, 2016 by Rebecca Hartman-Baker

There are upgrades to the electrical and plumbing systems in the NERSC machine room that must be performed in the new year. The electrical upgrade will increase the reliability of the electrical system that feeds power to the machines, and the plumbing system upgrade will increase the capacity of the water cooling system that keeps the supercomputers cool.

Read the full post

A Brief Primer on the SLURM Scheduler

October 17, 2016 by Rebecca Hartman-Baker

Edison is the only NERSC supercomputer currently available to users, so it is unusually oversubscribed. This has led to long wait times for many users, and a number of questions to NERSC consulting about the poor throughput many users are getting. Users often wonder whether the scheduler is working properly.

Read the full post