March 2015 - Acme-argonne - mailman3.cels.anl.gov

ACME Exascale Roadmap summit on Jan 23
by Robert Jacob 19 Nov '15

19 Nov '15

Hello everyone, FYI: The day after the 6-month review, a small summit is being held to: "Hold an initial discussion on climate-computing issues. Plan a larger workshop on strategies to maximize ACME performance and portability on current and future DOE LCFs." Paul Messina, will you be going? Agenda is attached. The white paper mentioned is just going to be a summary of past material on this issue and Mark Taylor will actually be writing it. Rob

3 3

6-month review
by Robert Jacob 12 Mar '15

12 Mar '15

ACME had a 6-month review on Jan 22nd in Germantown. This review was requested by the original proposal review committee. Review panel members are listed below. Agenda is attached. The ACME council attended along with a few group leads. Various BER program managers were in attendance besides Dorothy (Gary (in and out all day), Renu and Jay (present for most of review), Ashley and Sally (late-morning only)). Randall Lavolette from ASCR was also there. Like the Advisory Committee, the review panel was mostly complementary and we "passed". I haven't seen the written report yet but don't think it will prompt any changes by itself. The meeting ended early and the closed session on the agenda was skipped. From my notes (which emphasize the questions and negative points): - The review panel also thought biogeochemistry seemed out-of-place and we should focus on high-res, water-cycle and cryosphere. (ACME is not about to drop BGC. This is seen as a problem in how we explain the project.) - Ensembles were again mentioned as needing more attention from ACME - Balaji thought we needed more plans on fault-tolerance - Frank thought our hybrid computing strategy wasn't clear - Graeme wanted better metrics to show MPAS is an improvement - Elana felt the workflow/viz tools will need training and support I was again a little disappointed the panel didn't say "Performance and SE are hugely under-staffed!" There were no questions after my presentation except from Elena who wanted to be sure we can still do standalone runs of the atmosphere or land (we can). Let me know if you have any questions, Rob 6-month review panel: Bert Semtner, NPS Bill Tang, PPPL V. Balaji, GFDL Graeme Stephens, JPL Frank Giraldo (on phone), NPS Elena Shevliakova, Princeton

3 4

Advisory Committee meeting
by Robert Jacob 04 Mar '15

04 Mar '15

(First in a series of overdue updates) The ACME External Advisory Committee met in Denver on Jan 14th. This was the first meeting. Members are listed below. There were no fireworks. The highlights from their written report: - Impressed with progress so far. - Goals still seem to broad. Maybe focus on just water cycle and cryosphere. Need a better plan for handling risk. - Not enough emphasis on ensembles. - Wants to see more ACME staff approaching 100% on ACME. - ACME's target resolution of 10km ocean/25km atmosphere won't be world-leading in 3 years. Will have to lead on robustness. - Should consider 1km non-hydrostatic atmosphere for future. From my notes: - In their view, this is a separate model. Not buying the "its a branch of CESM" line. - Project has "fantastic" people and skills. - Sea level rise will be a clear winner for ACME. - Need to better coordinate with outside science developers. - Need to improve decision process. - Effort looking at coupled model is light. There were pretty silent on performance and software engineering probably because the more technical members weren't there. Let me know if you have any questions. Rob Advisory Committee Members: John Petch (chair) - UK Met Office Tom Delworth - GFDL Jeff Kiehl - NCAR David Holland - NYU Minghua Zhang - Stonybrook not present but joining the next meeting: V. Balaji (GFDL) and Paul Fischer

3 3

ACME-ASCR Exascale Summit group meeting
by Robert Jacob 04 Mar '15

04 Mar '15

The ACME-ASCR Exascale Summit Group meeting was held Jan 23rd at Germantown the day after the ACME 6-month review. The purpose of the meeting was "to figure out how facilities and research programs can collaborate with ACME on exascale and how to move forward." The main outcome of the meeting was agreement that they should work together and could move forward with another whitepaper. Dave Bader was pretty happy about not needing another workshop. He and Mark Taylor are supposed to write a white paper but I haven't seen any work on it. Binkley could not make it until about 12:30pm because of an exascale presentation. He was called out of the room a few times after. Randall Lavolette and Barb Holland attended the entire meeting for ASCR. Kathy Yellick, Paul Messina and Jim Hack represented the facilities. Dorothy Koch and Gary Geernart represented BER. From ACME: myself, Dave Bader, Pat Worley, Kate Evans, Andy Salinger, Bill Collins, Phil Jones, Mark Taylor and Esmond Ng. Dorothy emphasized that BER wanted to find new ways to partner with ASCR in meeting ACME's exascale goals. SciDAC is viewed as targeting the current machines. Barb was supportive. Details about what a new program would look like were not discussed. For me, the biggest surprise of this meeting was how prominently the "super-paramaterized" version of CAM was featured in Mark Taylor's presentation on ACME's and Exascale. SP-CAM adds a 2D high-res non-hydrostatic cloud-resolving model to each grid point of the course, hydrostatic parent model and improves many features of the simulated climate. It also makes CAM a better GPU candidate because the SP routine takes up 95% of the runtime providing a clear kernel to move to GPUs. It's been developed by Dave Randall at CSU but was not included in the ACME proposal. Not sure how/when exactly SP-CAM would be brought in to ACME. Mark outlined 3 elements for ACME to get to Exascale: high resolution non-hydrostatic global model, super-parameterization and ensembles. High-res only (without SP) is still needed because there are limits to what SP can do (there is no communication between SP instances on each course grid point and it makes no sense to use it with a high-res non-hydrostatic global model). Let me know if you have any questions. Rob

2 3