Harvard Uses Google Cloud to Clone Supercomputer for Medical Research Runs

Print Friendly, PDF & Email

Harvard University

A Harvard scientist used Google Cloud Platform compute resources to construct an HPC clone to conduct heart disease study, according to a Reuters story, “a novel move that other researchers could follow to get around a shortage of powerful computing resources and speed up their work,” the news service reported.

The study by Harvard’s Prof. Petros Koumoutsakos simulated the effects of a therapy designed to dissolve blood clots and tumor cells in the bloodstream. The compute-intensive study required supercomputing-class resources, Koumoutsakos told Reuters, explaining “the big problem that we had (was) we could run one simulation using a full scale supercomputer.” Modifying or optimizing the simulation required additional runs.

“Folks are realizing the potential for cloud to solve problems and technical scientific engineering computing to really unlock productivity and get to better answers, better insights, faster,” said Google Cloud’s Chief High Performance Computing technologist Bill Magro (formerly an Intel fellow and chief HPC technologist).

Cloud computing resources like Google’s are generally designed for small computing tasks for such workloads as streaming video, serving web pages or accessing databases. Getting clouds to behave like a supercomputer for scientific research on the lines of Koumoutsakos’s project, Magro told Reuters, requires modifications to cloud software, networking and physical design.

The project was conducted by Harvard, Citadel Securities and Google Cloud to demonstrate the validity of a novel, minimally invasive approach to unclogging arteries via a large-scale numerical simulation powered by the public cloud. By leveraging HPC running on cloud infrastructure, Harvard researchers hope to solve some of the world’s most complex calculations and advance heart disease research, the university said.

Prof. Koumoutsakos, leader of the research, is at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) and has worked on creating digital twins for therapeutic devices in circulatory diseases. Citadel Securities and Google Cloud are jointly funding work to demonstrate the power of scaled complex simulations for medical research.

“Today, the level of computing power required to run simulations of this complexity and at this scale is available from only a handful of supercomputers around the world,” said Prof. Koumoutsakos in an announcement issued by Harvard. “With the support of Citadel Securities and Google Cloud, we aim to demonstrate that public cloud resources can be harnessed to handle large-scale, high-fidelity simulations for medical applications. In doing so, we hope to show that easy access to massively available cloud computing resources can significantly reduce time to solution, improve testing capabilities and reduce research costs for some of humanity’s most pressing problems.”

In initial tests involving thousands of virtual machines spun up on Google Cloud, Prof. Koumoutsakos and his team, which also includes researchers from ETH Zurich, achieved 80 percent of the efficiency available in dedicated supercomputer facilities with extensively tuned code, according to Harvard. Two types of codes have so far been deployed on both GPU- and CPU-based architectures: a code that was a finalist for the Gordon Bell prize in 2015, and an open-source commodity code (LAMMPS) for particle simulations.