ISC Beyond the Hans Meuer Era
Hans Werner Meuer (1936 – 2014†) and his legacy need little introduction within the high-performance computing (HPC) community. In Europe, he is known as the “Father of European supercomputing.” Hans, as he was fondly known in the community, became involved in data processing in 1960 and for the next 54 years of his life, he played various roles in the supercomputing world.
While acting as the director of the computing center and professor for computer science, in 1986, Hans organized the world’s first supercomputing conference – the Mannheim Supercomputer Seminar. It its inaugural year, it drew 81 attendees. For a long time, this modest-sized conference was the only event for supercomputer manufacturers to exhibit their products and for users to discuss their applications.
Fifteen years later Hans changed the name to the International Supercomputing Conference, ISC, and establish a more professional tone by moving the conference to modern venues within Germany. He also added an exhibition that brought in key industry players as sponsors.
It was during the 1986 conference, with the help of a young colleague, Erich Strohmaier, that Hans started the “game” of publishing systems of the major supercomputer manufacturers of the day. At first, the list consisted only of the systems of vendors who attended the conference, regardless of relative compute speed. But due to the enormous performance difference between low-end and high-end models, the increasing availability of massively parallel processing (MPP) systems, and the strong increase in computing power of the high-end models of workstation suppliers, Jack Dongarra stepped in to help Hans add more structure to the list. In 1993, they started ranking the world’s most powerful computers according to the Linpack benchmark, a “standard” newly developed by Dongarra.
Hans’ inclination to continually refine whatever he was working on and explore new areas did not go unnoticed. Long-time friend and ISC program advisor, Horst Gietl was struck by Hans’ intellectual curiosity, recalling that he had little patience with conservative approaches. “He drew pleasure from discovering things for himself,” noted Horst. “He never was a follower of opinions of other people.”
It was this curiosity that led Hans to constantly inject new topics into the ISC program, attracting an ever-wider array of people. Today the event boasts 2,500 attendees. In five days, 300 speakers would present their topics of interest across 30 different sessions that run in parallel. Regular ISC attendees are surely able to recall Hans’ perennial opening quote, “This is our best conference ever!”
The tale of how Hans and Horst came to work together is an interesting one and exemplifies Hans’ ability for picking friends. Horst was an attendee of the Mannheim conference but they lost touch after he went to work on parallel processing systems used for video streaming. In 2006 they ran into each in Horst’s hometown, Munich, crossing paths for less than 10 seconds. Hans made it a point to reestablish the lost connection which later gave birth to a fruitful collaboration between them.
Some people claim that ISC will not be the same without Hans. His son Thomas agrees that his father’s winning personality will be sorely missed at ISC. But he noted, more than that, Hans was the figurehead. At the same time, ISC has evolved into a large multi-faceted event that takes scores of people to design the program and running the conference.
Recalling the amazing support, and condolences that poured in after Hans’ passing, Thomas is optimistic that the community will continue to see ISC as a significant HPC conference. “Many of our customers and attendees have personally conveyed to us that ISC is a must-attend event for them,” he said. “The reason is quite simple. We offer a quality program and also ample networking opportunities.”
That’s not to say the conference will remain static. Horst remarked that this year’s program is more application-oriented than in previous years. A number of sessions will focus on the “the real value of HPC,” theme including topics such as visualization, HPC in life sciences, extreme computing challenges, cloud computing, and trends for big data in HPC. There will also be a session in the industry track that discusses support structures for HPC in commercial enterprises.
Nevertheless, Horst noted that the 2014 program will also offer many sessions on more traditional supercomputer topics like programming models, future supercomputer directions, quantum computing, fault tolerance and resilience, performance measurement tools and power challenges.
“I would say the ISC’14 program is an interesting mix of HPC topics, which we hope will motivate the supercomputer community to join us in Leipzig,” said Horst.
The shifting balance between commercial and non-commercial HPC will be reflected in the conference. Over the next five years, he expects industrial HPC to gain more “attention” in the program. ISC introduced the industry track in 2013 and this year the focus is on commercial innovation via HPC technologies. Horst is hoping to see more simulation engineers and independent software vendors attending the two day program. “It is a known fact that the HPC requirements for the industry will grow for them to stay competitive in a globalized world …life science and finance are some examples,” explained Horst. The same is also true for social networks that require large HPC-systems to extract and analyze valuable information from the rapidly growing data volumes.
Another ISC change on the horizon is the conference venue. In 2015, which will mark ISC’s 30th anniversary, the event will be held in Frankfurt. For logistical reasons it will take place from July 12-16, breaking from the tradition of hosting it in June. According to Thomas the city is perfect for the next ISC. It offers a very modern convention facility, a huge range of hotels, perfect transportation (hint: Frankfurt Airport) and a vibrant downtown area within walking distance.
Asked about what will be new in 2015, Thomas was willing to offer this bit of information: “Since Frankfurt is one of the world’s most important financial centers, at least one session will be dedicated to financial services and its use in HPC cloud services. Furthermore we will be extending our scientific program and for the first time we will be offering a workshop day.”
Bernd Mohr, a senior scientist at the Juelich Supercomputing Centre whose work focuses on performance analysis of parallel software, joined the ISC program team a couple of months back as the future ISC Workshop Chair. Questioned on the need for a full day devoted to workshops in 2015, he explained that neither the conference session chairs nor presenters like having interesting workshops competing with their audience. “While they feel that workshops are stealing their audience, workshop organizers feel that they need to compete with the conference program, and the attendees complain that there is too much going on in parallel,” said Bernd. “Workshops will be ideal for those who always thought BoF sessions are too short to present and discuss their proposed topic,” he continued. “They are also ideal for European and international research projects that want to present their research results to a larger audience.”
Bernd is no stranger to the HPC scene or the ISC conference. He has spoken a number of times in the main conference session. “The Future of Performance Optimization Tools” was the topic of his first invited talk, which he admits he almost messed up because he was so nervous standing on the stage before the audience. Over the years he gradually became more involved in the conference as Hans noticed Bernd’s enthusiasm as a presenter and his strong interest in the conference. Bernd initially advised Hans and Horst as a freelance consultant, and in 2012 Hans offered Bernd a permanent post since he wanted him on the program team to improve the quality of the technical program.
Because of the July date for ISC’15, it looks like TOP500 fans will have to wait an extra month for the announcement of the 45th edition of the list in 2015! Regardless of the timeframe, the Linpack benchmark is often a subject of controversy in the community and undoubtedly this will continue to be the case in 2014, 2015, and beyond.
While the Linpack benchmark may not be as relevant in the much more diverse HPC application landscape that exists today, Thomas maintains that the simplicity of the list is its biggest advantage. “My assumption is that the current metric will continue to be the leading benchmark for the TOP500 list,” he said. “Although other benchmarks might be better suited for specific application problems, Linpack provides one single number and is easy to understand.”
According to Bernd, there is also the historical value of the list, which has provided a consistent way of measuring computer performance for over 20 years. As a result it has been invaluable in predicting the performance trajectory of supercomputers and analyzing architectural trends. But he also confesses that the Linpack benchmark is too FLOPS-centric and misses other important aspects of current HPC applications: “For me, the real value of HPC are the applications using these systems that solve real-world societal problems and these results and successes are not emphasized enough.”