A collaboration between the Department of Energy’s National Energy Research Scientific Computing Center (NERSC), Intel and five Intel Parallel Computing Centers (IPCCs) has resulted in a new Big Data Center (BDC) that will work both on code modernization and tackle real science challenges.
According to Prabhat, BDC Director and Group Lead for NERSC Data and Analytics and Services team, “The goal of the BDC is to solve DOE’s leading data-intensive science problems at scale on the NERSC Cori supercomputer. The BDC, in collaboration with Intel and the IPCCs, will test to see if the current HPC systems can support data-intensive workloads that require analysis of over 100 terabytes datasets on 100,000 cores or greater. The BDC will optimize and scale the production data analytics and management stack on Cori.
“Our first task will be to identify capability applications in the DOE data science community, articulate analytics requirements and then develop scalable algorithms.” Prabhat continued. “The key is in developing algorithms in the context of the production stack. Our multi-disciplinary team consisting of performance optimization and scaling experts is well positioned to enable capability applications on Cori. All the optimizations done at the BDC will be open source and made available to peer HPC centers as well as the broader HPC and data analytics communities.”
Quincey Koziol, BDC co-director and principal data architect at NERSC, noted, “While data analytics is undoubtedly the rage at this point in time, scalable analytics fundamentally relies on a robust data management infrastructure. We will be working on examining the performance of parallel I/O as exercised through modern data analytics languages (Python, R, Julia) and machine learning/deep learning libraries.”
Joseph Curley, director for Intel’s code modernization efforts, states, “We’ve combined the BDC goal of providing software stacks and access to HPC machines where the data driven methods can be developed with our IPCC program. We married the two ideas together by combining research members of the community with a program that we have for outreach.
“The objective of the Big Data Center (BDC) comes from a common desire in the industry to have software stacks that can help the NERSC user base, using data driven methods, to solve their largest problems at scale on the Cori supercomputer. So one of our main goals is to be able to use the supercomputer hardware to its fullest capability. Some underlying objectives at BDC are to build and harden the data analytics frameworks in the software stack so that developers and data scientists can use the Cori supercomputer in a productive way to get insights from their data. Our work with NERSC and the IPCCs will involve code modernization at scale as well as creating the software environment and software stack needed to meet these needs.”
The five IPCCs who are part of the BDC program include the University of California-Berkeley, University of California-Davis, New York University (NYU), Oxford University and the University of Liverpool. Their initial BDC work includes this research:
- The University of California-Berkeley team is working on the Celeste project. Celeste aims on developing highly scalable inference methods for extracting a unified catalog of objects in the visible universe from all available astronomy data.
- The University of California-Davis group is working on development of computational mechanics techniques to extract patterns from climate simulation data. The techniques build upon techniques in information theory to achieve unsupervised pattern discovery.
- The New York University (NYU) team is working on extending deep learning to operate on irregular, graph-structured data. The techniques are being applied to problems in high-energy physics.
- The Oxford University group is developing a new class of methods called probabilistic programming and applying the methods to challenging pattern and anomaly detections in high-energy physics. The work combines probabilistic programming with deep learning to train large networks on Cori.
- The University of Liverpool team is working on developing topological methods to analyze climate datasets. The techniques are being used to extract stable, low-dimensional manifolds, and robust pattern descriptors for weather patterns.
The BDC work will also benefit the larger data analysis and HPC communities. Curley states, “In our IPCC program, we encourage system users to discover methods of solving problems on HPC systems, document what they did, and teach others how to follow their methods. This creates a beneficial cycle of new research, hardening the machine, developing new software stacks leading to research that is more productive—what we call a virtuous cycle.
“We are excited about the IPCCs working in conjunction with the BDC because there will be people working on problems we never could have anticipated and advancing human knowledge in ways we never could have guessed. If you can combine this with the BDC at NERSC that has a large machine like Cori and a diverse user group, you end up creating networks of knowledge and creating scientific results that are unpredictably wonderful.”
About the Author
Linda Barney is the founder and owner of Barney and Associates, a technical/marketing writing, training and web design firm in Beaverton, OR.
Feature image: Berkeley Lab Shyh Wang Hall, home of NERSC