The paradigm of high-performance computing has flipped: “super” computers, with their thousands of custom processing cores are now no match for the might of the cloud – hundreds of thousands of cores available on demand and highly scalable. At the same time, many large scale, complex computer simulations that are tackling many of humankind’s toughest scientific challenges were specifically optimized for running in these less optimal supercomputing environments. With computer simulations growing increasingly complex, this limitation is standing in the way of the next level of scientific advancement in fields such as fusion energy, climate change modelling, computational biology and many others.
We want to change the game and run these powerful simulations on standard industrial clusters. We are seeking extreme scalability through things like algorithm choices, alternate parallel methods, automatic parallelization or others we haven’t even dreamed of: at its ‘core’ we are seeking intelligent algorithms to allow a revolution in scientific computing. Ultimately we seek highly scalable scientific computing in the cloud that is easier to access and dramatically less expensive.
Many scientific fields use large, complex and highly iterative computer simulations to advance knowledge in fundamental ways. Unfortunately, many of these codes were built with high-performance computing/supercomputers in mind, or they have recently become large enough that they need to make the transition to parallel computing. This often requires physicists, biologists, and the scientific community to engage in the complexities and nuances of how to parallelize and optimize code instead of the science itself. We envision a world in which scientists can focus on science and have their critically important simulation codes run on standard industrial cluster environments in a dramatically simpler manner, at higher fidelity, exponentially faster and cheaper. Industrial clusters provide massive hardware computing resources, and we want to incentivize developing code to better use those resources. Today, the advancement of many scientific disciplines are limited by the difficulties of parallelizing code.
As an example, Computational Fluid Dynamics (CFD) codes are used to simulate a broad range of systems that behave like fluids – everything from fusion energy through to climate change. These complex computer models could be the key to unlocking some of the biggest questions of contemporary life: clean energy solutions, advanced cures and preventative medicine, and global predictions for climate change. The majority of these complex simulations, however, are only run in supercomputing environments. This limitation restricts the rate at which scientific advancements can be accomplished. The problem persists well beyond CFD codes however.
The world has seen an explosion of cost-effective cloud computing environments that now vastly outmatch the processing power of supercomputers. Amazon Web Services and Google Cloud Platform, among others, offer thousands of times the processing power of large supercomputers. Ironically, critical advances in fusion energy, climate modeling and many other fields are now limited by their inability to run as efficiently on these more standard cluster environments.
Example: Importance of computer simulation in fusion energy: Fusion power could be a permanent solution to the world's energy needs. Fusion energy has no harmful effects on the environment, no risk to human life through meltdowns or proliferation, and planet earth has millions of years of fusion fuel already. We know fusion is possible because it is the same process by which stars generate heat.
Fusion experiments are often expensive and time-consuming, which incentivizes the use of computer simulations to help us understand how well a real-life experiment would work. The supercomputing resources required to run current fusion models, however, hinders the ability to engage the critical mass of researchers trying to solve these problems. Enabling them with faster and easy to access computing resources will expedite breakthroughs.
Though the science involved in fusion energy, weather and climate modelling, and computational biology are very different, there is plenty of room for optimizing how the codes run on industrial clusters. The challenge is the same across these and many other fields.
But we don’t want to build more supercomputers. We want to change the game.
Simply put, we want to run these powerful simulations on industrial clusters. The computing power available through these clusters is enormous compared to current high-performance supercomputers, and we want to tap into this potential.
We are reaching out to you, a problem-solving community of experts around the world, to tackle this problem. If we were able to run highly iterative simulation codes quickly on standard clusters, scientific computing as a whole could undergo a revolution.
The solutions to this problem could come in a variety of forms, from new algorithms to automatic parallelization, to hardware, and everything in-between. We need your bold, brash, brazen ideas to bring about this solution. The benefits this breakthrough would bring to all of humanity are difficult to underestimate.
To achieve this breakthrough, we need to be able to run a wide range of different codes on an industrial cluster environment such as Amazon Web Services or Google Cloud Platform. We are continuing to develop the required benchmarks and milestones required to win the prize and would like your engagement to help do so. So far, we know we want to focus at least the following items: cost per simulation, and length of time to run a simulation.