This week the tenth biennial International Advanced Research Workshop on High Performance Computing, HPC2010, is taking place, as with the previous nine, at the Grand Hotel San Michele in Cetraro, Italy. Cetraro is a small town on the western coast of Italy, about two hours south of Naples by rail. The workshop, hosted by the University of Calabria, was originally focused on high-performance computing, but that focus has expanded over the years to include grid computing, and this year further expanded to “High Performance Computing, Grids and Clouds.”
Professor Lucio Grandinetti from the University of Calabria (in Cosenza, about 45 minutes inland) welcomed this year’s 110 participants, from 27 countries with a retrospective on changes in the industry and the research community in the nearly two decades since the first workshop was held in 1992. The demographics for this invitation-only gathering are typical this year, with about half of the participants from universities, 30 percent from national laboratories, and 20 percent from industry. The conference relies on a growing team of now 27 of sponsors.
Days one and two included lectures from 22 attendees, many of which were motivated by the exascale goal. In each session there were four to six talks followed by a Q&A session with the speakers, culminating with an evening panel on “Challenges and Opportunities in Exascale Computing,” chaired by Dr. Paul Messina (Argonne National Laboratory) and including eight experts from industry, laboratories and universities. One of the themes related to exascale was that many necessary advances can be achieved in an evolutionary fashion, but these as well as more fundamental breakthroughs will rely on a much more deliberately coordinated set of partnerships than have been seen in the past. These partnerships, for example the International Exascale Software Project (IESP), will cut across labs, universities and industry as well as involving “co-design” strategies where hardware, system software, and application experts work together. The need to carefully craft necessary intellectual property, licensing and support agreements to these partnerships was noted by several participants.
Panelists and speakers also stressed the need to be opportunistic in leveraging innovations emerging from the marketplace. The IESP software roadmap takes this concept forward in the form of a 2-by-3 classification matrix guiding prioritization. The matrix relates critical versus non-critical needs to three levels of applicability: unique to exascale, potential for benefit to current scale (tera-, peta-), and commodity.
The evening concluded with the third consecutive installment of the “singing panel,” now a tradition at the HPC workshop, mixing the melodic voices of computer scientists with a chorus of iPhone Ocarina instrumentalists. Past panels sang about open source (“With Some Middleware from My Friends,” to the tune of the Beatles’ “With a Little Help from my Friends”) and cloud computing (“Get Off My Cloud,” to the obvious Rolling Stones tune). This year the exascale panel and past-panel alumni sang of exascale challenges in an adaptation to the Rolling Stones’ “You Can’t Always Get What You Want.” A video release is rumored, though Dan Reed (Microsoft) wisely advised all singing panelists early in the tradition to “keep your day jobs.” On this point there was strong community consensus.