Measuring What Matters
Turing Award Winner Jack Dongarra Shines a Light on the State of Supercomputing
Today at SC’22 the Turing Award winner, Jack Dongarra of the University of Tennessee, provided a retrospective on high performance computing from its early days to opportunities of the Exascale era. A note on the Turing Award. Named after Alan Turing and often considered a Nobel equivalent for computing, the award goes to a scientist each year who has contributed to the field of computing advancement and comes with a $1,000,000 prize. Past recipients have delivered inventions like the UNIX operating system (Ken Thompson and Dennis Ritchie in 1983), TCP/IP (Vint Cerf and Bob Kahn in 2004), compiler and automatic parallel execution (Frances Allen, the first woman awarded the Turing in 2006), and design of computer architectures (David Patterson and John Hennessy in 2017). These are people who can hush rooms when speaking, and this year’s award recipient is no different.
One could argue that the HPC field and SuperComputing conference itself would be vastly different without Jack Dongarra’s contributions. His continuous investment in eras of HPC development have enabled the scientific community to both maximize the value of infrastructure and measure infrastructure performance. It’s this last bit that offered a unique lesson for those listening to Jack’s talk today. It’s the story of how the Top 500 came into being and really how we have standard benchmarks for everything the computing industry delivers. This, for someone like me who has been in this industry for a minute or two, is like uncovering why water is wet. We care deeply for advancements in performance, and things like benchmarks and the Top 500 give us the fundamental tools to measure advancements in a fair and consistent manner.
So how did this happen? Jack, as one of the founders of Linpack, wrote a simple table to measure a set of complex equations to measure relative performance of systems back in the 1980s. He began maintaining a list of relative performance, and in 1993 merged his efforts with two other scientists’ similar pursuits. The Top 500 was born with a system from Los Alamos Labs at the top measuring a total of 1000 processors busy simulating nuclear warheads. Since then labs vie for top spots on the Top 500, vendors scramble to ensure their infrastructure is featured prominently, and the entire computing industry is pushed forward reaching for more performance to fuel these massive machines. This is a great example of the human motivation driven in part by standard metrics, and we all have to thank Jack and his colleagues for this bar.
So what has happened since the introduction of the Top500? Jack spoke about waves of computing architectural transformations, from shared memory systems in the ‘90s to distributed memory systems of the 2000’s to the introduction of multicore and hybrid architectures in the 2010’s and today the era of exascale fueled by the merger of HPC and AI, the evolution of heterogeneous platforms and ultimately the performance at Exascale (or a billion billion flops). The Exascale Project, with multi-billion dollar funding by the US government, has focused on 21 applications across the realms of scientific exploration with a common theme of 3D model simulation, all benefited by underlying infrastructure advancements and software optimization efforts and contributions from Dr. Dongarra. We’ll continue to see the secrets of the universe unlocked at more rapid rates due to the collective contributions of this HPC community, and for that I’m incredibly grateful.
Still, there are some challenges ahead. The Top 500 has also become a geo-political hot potato. With China holding the pole position with over 160 supercomputers on the list and the US in second position with 125 there is an existential reality that those who can unlock these secrets have power of first action. In fact, there is belief that China holds two Exascale class machines that it has not submitted to the Top 500 list in this era of microprocessor power becoming a strategic national security asset. One need not look farther than recent history of China and technology sanctions and the US CHIPS Act to understand the value of microprocessors to national security interests, and the nexus for this challenge is playing out on the Top 500 list. And in reflecting on this, I can only think back to yesterday’s plenary session at the conference where leaders of supercomputing from ORNL and NIAID called for a new era of collaborative research to solve our most pressing problems. This open and collaborative approach proved to be critical most recently to unlocking the COVID 19 virus and has been a cornerstone of computer aided research since its inception. In this way, the Top500 is providing a different kind of transparent measurement, one that shines a light on how geopolitical concerns can stand in the way of societal advancement. I’m hoping that the spirit of Jack Dongarra and his fellow Turing Award winners prevail to all of our benefit. Thanks for engaging. - Allyson
Turing Award Winner Jack Dongarra Shines a Light on the State of Supercomputing
Today at SC’22 the Turing Award winner, Jack Dongarra of the University of Tennessee, provided a retrospective on high performance computing from its early days to opportunities of the Exascale era. A note on the Turing Award. Named after Alan Turing and often considered a Nobel equivalent for computing, the award goes to a scientist each year who has contributed to the field of computing advancement and comes with a $1,000,000 prize. Past recipients have delivered inventions like the UNIX operating system (Ken Thompson and Dennis Ritchie in 1983), TCP/IP (Vint Cerf and Bob Kahn in 2004), compiler and automatic parallel execution (Frances Allen, the first woman awarded the Turing in 2006), and design of computer architectures (David Patterson and John Hennessy in 2017). These are people who can hush rooms when speaking, and this year’s award recipient is no different.
One could argue that the HPC field and SuperComputing conference itself would be vastly different without Jack Dongarra’s contributions. His continuous investment in eras of HPC development have enabled the scientific community to both maximize the value of infrastructure and measure infrastructure performance. It’s this last bit that offered a unique lesson for those listening to Jack’s talk today. It’s the story of how the Top 500 came into being and really how we have standard benchmarks for everything the computing industry delivers. This, for someone like me who has been in this industry for a minute or two, is like uncovering why water is wet. We care deeply for advancements in performance, and things like benchmarks and the Top 500 give us the fundamental tools to measure advancements in a fair and consistent manner.
So how did this happen? Jack, as one of the founders of Linpack, wrote a simple table to measure a set of complex equations to measure relative performance of systems back in the 1980s. He began maintaining a list of relative performance, and in 1993 merged his efforts with two other scientists’ similar pursuits. The Top 500 was born with a system from Los Alamos Labs at the top measuring a total of 1000 processors busy simulating nuclear warheads. Since then labs vie for top spots on the Top 500, vendors scramble to ensure their infrastructure is featured prominently, and the entire computing industry is pushed forward reaching for more performance to fuel these massive machines. This is a great example of the human motivation driven in part by standard metrics, and we all have to thank Jack and his colleagues for this bar.
So what has happened since the introduction of the Top500? Jack spoke about waves of computing architectural transformations, from shared memory systems in the ‘90s to distributed memory systems of the 2000’s to the introduction of multicore and hybrid architectures in the 2010’s and today the era of exascale fueled by the merger of HPC and AI, the evolution of heterogeneous platforms and ultimately the performance at Exascale (or a billion billion flops). The Exascale Project, with multi-billion dollar funding by the US government, has focused on 21 applications across the realms of scientific exploration with a common theme of 3D model simulation, all benefited by underlying infrastructure advancements and software optimization efforts and contributions from Dr. Dongarra. We’ll continue to see the secrets of the universe unlocked at more rapid rates due to the collective contributions of this HPC community, and for that I’m incredibly grateful.
Still, there are some challenges ahead. The Top 500 has also become a geo-political hot potato. With China holding the pole position with over 160 supercomputers on the list and the US in second position with 125 there is an existential reality that those who can unlock these secrets have power of first action. In fact, there is belief that China holds two Exascale class machines that it has not submitted to the Top 500 list in this era of microprocessor power becoming a strategic national security asset. One need not look farther than recent history of China and technology sanctions and the US CHIPS Act to understand the value of microprocessors to national security interests, and the nexus for this challenge is playing out on the Top 500 list. And in reflecting on this, I can only think back to yesterday’s plenary session at the conference where leaders of supercomputing from ORNL and NIAID called for a new era of collaborative research to solve our most pressing problems. This open and collaborative approach proved to be critical most recently to unlocking the COVID 19 virus and has been a cornerstone of computer aided research since its inception. In this way, the Top500 is providing a different kind of transparent measurement, one that shines a light on how geopolitical concerns can stand in the way of societal advancement. I’m hoping that the spirit of Jack Dongarra and his fellow Turing Award winners prevail to all of our benefit. Thanks for engaging. - Allyson
