Authors: Peter Beckman (Argonne National Laboratory, Northwestern University), Mark Asch (University of Picardie), Jack Dongarra (University of Tennessee)
Abstract: The emergence of data-intensive machine learning and AI in a wide variety of scientific fields, and the explosive growth of data generated by new instruments and IoT in edge environments, is severely disrupting the landscape for scientific computing. The international Big Data and Extreme-scale Computing (BDEC) initiative has started a second workshop series—BDEC2—with a formulate a coordinated international response to these powerful trends. This BoF will review the results of the three workshops held 2018-2019 and seek input from the SC community into the next three workshops, which aim to deliver a candidate software infrastructure roadmap in 2021.
Long Description: Over the past six years, the Big Data and Exascale Computing (BDEC) community has carried out a series of international workshops (funded by the NSF, DOE, EU, Japan and China) to develop a plan for transnational cooperation in the design and development of a next generation software infrastructure for extreme scale and data driven science and engineering. The first series of workshops (2013-2017) produced the “Pathways to Convergence” report, which was presented at SC17 (https://www.exascale.org/bdec/report). It highlighted two areas of highly disruptive change that any roadmap for future cyberinfrastructure must take into account: 1) the torrential increases in the volume and variety of data flowing from a myriad of devices, in a huge range of sizes and types, and spread across the “computing continuum,” from cloud and HPC data centers to smart infrastructure, new instruments and IoT at the network’s edge; and 2) the explosive growth in the use of data-intensive machine learning and artificial intelligence to harness these huge new data inflows for science and engineering.
Since these two trends are major drivers of massive AI research and development initiatives that are currently ramping up in the United States and other countries, the urgency of finding a path to a common software ecosystem to enable next generation science and engineering has increased substantially. Accordingly, starting in the fall of 2018, the BDEC community staged the first three in a second series of six more workshops (hence, BDEC2) designed to clarify and make more concrete a plan for a software infrastructure for data-intensive, AI-enabled science and engineering. Focusing on application requirements (Bloomington, IN), platform design (Kobe, Japan), and prototype demonstrations (Poznań, Poland), these three workshops greatly improved our understanding of AI-enabled application requirements and of the need for distributed services to support workflows stretching across the computing continuum. But the work of converging on a distributed platform design, identifying key holes in the software environment, finding needed integration points, and fostering the requisite community coordination are far from finished.
The proposed BDEC2 community roundtable at SC19 aims to gather important community responses to the work done in the first three BDEC2 workshops, and input into the next three workshops, which aim to deliver a candidate software infrastructure roadmap in 2021. BoF leaders will seek feedback from attendees on the emerging plan for a software infrastructure that can integrate scientific computing and HPC into the new world of data-intensive AI methods and into the use of the huge panoply of sensors and computing devices that define the emerging computing continuum.
The format will be two or three presentations from BDEC2 leadership on the outcomes of the previous meetings, including a summary presentation on the proposed “demonstrator” projects and targeted application domains, followed by community discussions and presentations. The latter will be open to BOF participants but primed by a few selected invited expert presentations. The international contribution and themes of this BDEC effort intersect with an extremely broad cross section of interests of the SC’19 community.
Back to Birds of a Feather Archive Listing