JUQUEEN Extreme Scaling Workshop 2017

JUQUEEN Extreme Scaling Workshops provide selected application teams the opportunity to scale their codes up to 1.8 million hardware threads via exclusive access to the entire Blue Gene/Q at JSC, with expert hardware, software, and support personnel available to assist with running codes at this extreme scale.

Following the positive feedback from the 2015 and 2016 workshops, where seven code teams qualified for membership of the High-Q Club, we are planning to host another three-day workshop from 23 to 25 January 2017 with sufficient time set aside for performance analysis of large-scale executions and quantification of scaling inefficiencies.

Participation will be limited to a small number of application teams with codes that demonstrate good scaling potential, and where at least one team member will be present on-site for the workshop.

While we are particularly keen to support new application teams and codes, teams from previous extreme scaling workshops and High-Q Club members with application codes that have already demonstrated excellent scalability are not excluded if algorithmic – or other – changes are being investigated or the focus is on improved file I/O, etc.

Participation is by invitation only, so please contact us with the details of your code, its current scalability on Blue Gene/Q, and the expected benefits from this opportunity to demonstrate extreme scalability (e-mail to Brian Wylie, b.wylie@fz-juelich.de, or Dirk Brömmel, d.broemmel@fz-juelich.de). We will follow-up to refine proposals and confirm availability, with selection continuing until all available places are filled. You should therefore contact us as soon as possible if you are interested.

Please note that there is a special queue for JUQUEEN for jobs with more than 16 racks, which are run following maintenance or on a weekly basis as required, as well as regular Big Blue Gene Weeks dedicated to running large-scale jobs. The Extreme Scaling Workshop is a complementary support offer, which can serve as valuable preparation for the effective utilization of JUQUEEN and future large-scale computer systems.
(Contact: Dr. Brian Wylie, b.wylie@fz-juelich.de)

JSC News No. 245, 18 November 2016