The University of Michigan is collaborating with IBM to develop and deliver “data-centric” supercomputing systems designed to increase the pace of scientific discovery in fields as diverse as aircraft and rocket engine design, cardiovascular disease treatment, materials physics, climate modeling and cosmology.
The system is designed to enable high performance computing applications for physics to interact, in real time, with big data in order to improve scientists’ ability to make quantitative predictions. IBM’s systems use a GPU-accelerated, data-centric approach, integrating massive datasets seamlessly with high performance computing power, resulting in new predictive simulation techniques that promise to expand the limits of scientific knowledge.
The collaboration was announced this week in San Jose at the second annual OpenPOWER Summit 2016. The OpenPOWER Foundation, which U-M recently joined, is an open, collaborative, technical community based on IBM’s POWER architecture. Several other Foundation members contributed to the development of this new high performance computing system, which has the potential to reduce computing costs by accelerating statistical inference and machine learning.
|OpenPOWER Foundation: new servers and big data analytics innovations|
|At the second annual OpenPOWER Summit, the OpenPOWER Foundation revealed more than 50 new infrastructure and software innovations, spanning the entire system stack, including systems, boards, cards and accelerators. These new products build upon 30 OpenPOWER-based solutions already in the marketplace.|
|Foundation members introduced more than 10 new OpenPOWER servers, offering expanded services for high performance computing and server virtualization.|
Working with IBM, U-M researchers have designed a computing resource called ConFlux to enable high performance computing clusters to communicate directly and at interactive speeds with data-intensive operations.
The ConFlux cluster will be built with ~43 IBM Power8 CPU two-socket “Firestone” S822LC compute nodes providing 20 cores in each, and fifteen Power8 CPU two-socket “Garrison” compute nodes providing an additional 20 cores each. Each of the Garrison nodes will also host four NVIDIA Pascal GPUs connected via NVIDIA’s NVLink technology to the Power8 system bus. Each node has a local high-speed flash memory for random access.
All compute and storage is connected via a 100 Gb/s InfiniBand fabric. The IBM and NVLink connectivity, combined with IBM CAPI Technology will provide an unprecedented data transfer throughput required for the data-driven computational physics researchers will be conducting.
Hosted at U-M, the project establishes a hardware and software ecosystem to enable large-scale data-driven modeling of complex physical problems, such as the performance of an aircraft engine, which consists of trillions of molecular interactions.
ConFlux, funded by a grant from the National Science Foundation, aims to advance predictive modeling in several fields of computational science. IBM is providing servers and software solutions.
There is a pressing need for data-driven predictive modeling to help re-envision traditional computing models in our pursuit to bring forth groundbreaking research. The recent acceleration in computational power and measurement resolution has made possible the availability of extreme scale simulations and data sets. ConFlux allows us to bring together large scale scientific computing and machine learning for the first time to accomplish research that was previously impossible.—Karthik Duraisamy, assistant professor in the U-M Department of Aerospace Engineering and director of U-M’s Center for Data-driven Computational Physics
ConFlux meshes well with IBM’s recent focus on data-centric computing systems.
Scientific research is now at the crossroads of big data and high performance computing. The explosion of data requires systems and infrastructures based on POWER8 plus accelerators that can both stream and manage the data and quickly synthesize and make sense of data to enable faster insights.—Sumit Gupta, vice president, high performance computing and data analytics, IBM
Advanced technologies such as data-centric computing systems are at the forefront of tackling big data challenges and advancing the pace of innovation. By moving computing power to where the data resides, organizations of all sizes can maximize performance and minimize latency in their systems, enabling them to gain deeper insights from research. These data-centric solutions are accelerated through open innovation and IBM’s work with other members of the OpenPOWER Foundation.
The incorporation of OpenPOWER technologies into a modular integrated system will enable U-M to configure the systems for their specific needs. ConFlux incorporates IBM Power Systems LC servers, which were designed based on technologies and development efforts contributed by OpenPOWER Foundation members including Mellanox, NVIDIA and Tyan. It is also powered by the latest additions to the NVIDIA Tesla Accelerated Computing Platform: NVIDIA Tesla P100 GPU accelerators with the NVLink high-speed interconnect technology. (Earlier post.)
Additional data-centric solutions U-M is using include IBM Elastic Storage Server, IBM Spectrum Scale software (scale-out, parallel access network attached storage), and IBM Platform Computing software.
In an internal comparison test conducted by U-M, the POWER8 system significantly outperformed a competing architecture by providing low latency networks and a novel architecture that allows for the integrated use of central and graphics processing units.
As one of the first projects U-M will undertake with its advanced supercomputing system, researchers are working with NASA to use cognitive techniques to simulate turbulence around aircraft and rocket engines.
They’re combining large amounts of data from wind tunnel experiments and simulations to build computing models that are used to predict the aerodynamics around new configurations of an aircraft wing or engine. With ConFlux, U-M can more accurately model and study turbulence, helping to speed development of more efficient airplane designs. It will also improve weather forecasting, climate science and other fields that involve the flow of liquids or gases.
U-M is also studying cardiovascular disease for the National Institutes of Health. By combining noninvasive imaging such as results from MRI and CT scans with a physical model of blood flow, U-M hopes to help doctors estimate artery stiffness within an hour of a scan, serving as an early predictor of diseases such as hypertension.
Studies are also planned to better understand climate science such as how clouds interact with atmospheric circulation, the origins of the universe and stellar evolution, and predictions of the behavior of biologically inspired materials.