Patrick Schmitz (Associate Director, Research IT & Program Manager, Berkeley Research Computing) will share his impressions of the CASC (Coalition for Academic Scientific Computing) Fall meeting, and the Supercomputing 2016 conference. BRC staff Krishna Muriki and Yong Qin will join Patrick in describing their impressions of Supercomputing 2016.
Please join us this Thursday, 1 December, to hear reportbacks from these two important national conferences:
Prior to the meeting, participants may wish to review the CASC agenda and Supercomputing agenda to get a sense of the scope of each conference. Suggested conference highlights for those who wish to dive more deeply:
Some CASC highlights of interest:
Presenting: Patrick Schmitz, Research IT; Krishna Muriki (LBNL & Research IT)
Aaron Culich, Research IT
Aron Roberts, Research IT
Barbara Gilson, SAIT
Chris Hoffman, Research IT
David Greenbaum, Research IT
Gary Jung, LBNL, Research IT
Greg Kurtzer, LBNL & Research IT
Jason Christopher, Research IT
John Lowe, Research IT
Jukka-Pekka Vainio, Berkeley Summer Session
Yong Qin, LBNL & Research IT
Kelly Rowland, Research IT
Maurice Manning, Research IT
Quinn Dombrowski, Research IT
[PLS Notes on Google Drive - Links; Krishna's notes -- COMING SOON]
CASC: higher-ed research computing conference ... org has a couple of directions, including meeting with program officers and leadership from national funder organizations (DOE, NSF) to learn about those organization's programs; and to meet with peers to discuss what's happening in research computing across EDU space.
trustedci.org -- Indiana U -- free help to institutions on many matters to do with cybersecurity, including grant proposal development.
Georgetown - risk analysis on data that requires security of any kind
Purdue - paper earlier this year, setting up secure data in AWS - $38K for a project to participate. Article worth reading.
Collaborative research a risk. Shadow IT orgs, e.g., fit-bit or other such data made public on a public site, many researchers are eager to use this; questions about what obligation (IRB, etc.) researchers and institutions have regarding research done on this data. Big issue in Europe.
NSF Future Directions Plan
On where to place this in NSF program structure
No solutions on roadmap for long term data storage
Concern about network not being acknowledged as a key infrastructure need in computation/data based research
Discussion about sys admin function of HPC moving to cloud and national centers; how that is distinct from need to support researchers in using HPC and cloud resources.
DH? NSF is clearly a science org, Kevin Thompson said. Breadth good. Reviewers are scientists.
ROI on consulting: "how long would it have gotten to get to the point you got if you had not been supported by XSEDE Champion" -- 15 mos researcher answer vs. 3 mos investment in XSEDE Champion, 5:1 ROI ... [PLS is thinking about how Research IT might use some question like this in our engagement, and integrate this into our own metric measurement]
Mobility of Compute - Dave Lifka at Cornell, "Project Aristotle"
JetStream still recruiting for allocations. 25% of VM capacity; 6% utilization. (Quinn notes that this may have something to do with interest in DH)
TACC -- training and courses to help people use computation in research. Created a department to award credit for courses offered.
Workshop on women in HPC (Patrick)
* As we are serving people outside pure sciences, we need a broader set of people who can engage with the researchers in those domains.
* Documented superior performance in creativity, patent awards, grant awards, etc. when a team is more diverse.
* Admonition to recruit all the time, not just when one has a job opening.
* Avoid gender and culture invisibility: a recipe for ignoring the problem.
* HPC for change. Kelly at TACC spoke about an event billed as this, very high attendance by woman and people of color. Interesting to think about how we frame why we do HPC, what we're offering
* Implicit bias online tests. Interesting on an individual level, but also interesting discussions can be had among people who take such tests
Software engineering for science (Patrick)
* Much improved from prior year, in which software engineering was seen as having little to no place in a research environment
* testing frameworks
* compiler settings can materially affect results
* building a shed (no need for an engineer/architect) vs. building a barn/building
* difference between researchers aiming to become an engineer or architect in software space; vs. consulting with someone who has applicable experience & skills
Krishna also attended the (overlapping) Intel HPC Developer Conference, at which a half-day Singularity session was led by Greg Kurtzer with Krishna's help.
Top HPC machine is 100 Petaflops, Chinese. Oak Ridge has top US machine at 17 PF
Various vendor observations, technology developments, details not noted here ... see Krishna's notes
Krishna's belief: KNL (Knight's Landing from Intel) and subsequent many-core CPUs from Intel, will eventually kill GPUs and HPC