BRC HPC Condo Cluster Service gaining broad adoption

Schematic overview of the Savio HPC cluster

Berkeley’s new High Performance Computing Condo Cluster has expanded rapidly in the first nine months of its operation, with hardware contributions to Savio from faculty in the physical, biological, and social sciences more than doubling the cluster’s initial size.

Researchers in a wide range of disciplines have joined the Savio Condo. Condo participants currently include research groups from Astronomy, Astrophysics, Atmospheric Sciences, Chemistry, Civil and Environmental Engineering, Earth and Planetary Sciences, Geography, Molecular and Cell Biology, Nuclear Engineering, Physics, Political Science, the Social Sciences D-Lab, and Statistics. The initial campus investment in Savio included purchase of 76 nodes; augmented by researcher contributions through early 2015, the cluster has grown to 172 nodes with a total of 3440 processor cores.

Recent condo contributor and Director of the D-Lab, Economist and Professor of Law Justin McCrary, shared his perspective on Savio shortly after beginning to use the new cluster:

"Every social scientist on campus feels the growing influence of large data collections on the scope and direction of social science. This avalanche of data is both a cause and a consequence of exponential growth in computing power and the concomitant growth in human activities taking place on the internet. It poses a major challenge to the university, but also presents an opportunity. As Director of D-Lab, one of my most important goals is to make sure that the BRC program's condo cluster succeeds and that faculty and students know how to use it effectively. I expect great things to emerge from Berkeley social science in the next five years, and I expect Savio to be a big part of that."

Integrative biology Professor Rasmus Nielsen described the effect he expects the BRC program to have on his colleagues’ research in a presentation at the Berkeley Research Computing launch in May 2014:

"Biology these days is being transformed into becoming much more computational. And many people in biology, in the biological community, ... don’t have access to the computing to deal with this. ... And that’s really where the BRC will make a huge difference for the biology community. ... I predict that the BRC will fundamentally transform how many people do science on campus."

At the same event, Professor David Culler of EECS described his vision of a flourishing of “real-time science” enabled by the “production capabilities” of teams like the Berkeley-LBNL collaboration that operates Savio. As an example of research in this vein, he cited the work of Ron Cohen of the Berkeley Atmospheric Sciences Center, which examines “mashups” of sensor data streaming from space satellites and earthbound sensors and requires a broad range of massively parallel computation over data representing fluid dynamics, chemical kinetics, particle density, and atmospheric radiation.

High Performance Computing on Savio Cluster

A High Performance Computing (HPC) cluster such as Savio supports research through computer modeling, simulation, and analysis, often over very large data sets. An HPC cluster is equipped with specialized, high-speed communication between racks of state-of-the-art server nodes; utilizes a parallel-access filesystem for speedy storage operations on data input and results; and runs software that distributes computation across the cluster’s multiple nodes to analyze and transform data in parallel, simultaneous streams of work. HPC is one of four principal aspects of the Berkeley Research Computing (BRC) Program, which also includes cloud computing, virtual workstations, and consulting.

Savio was launched in May 2014 as the initial phase of the BRC Program, funded by the offices of Vice Chancellor for Research Graham Fleming, Chancellor Nicholas Dirks, and Chief Information Officer Larry Conrad. Astronomer Eliot Quataert and nine colleagues in Astronomy, Earth and Planetary Science, Materials Engineering, and Physics sparked the formation of the BRC Program with their May 2013 proposal to VCR Fleming "that UC Berkeley adopt a condominium computing model."

In the “Condo” model, the campus covers the cost of key HPC infrastructure (power, cooling, high-speed networking, and a fast parallel storage array) and provides system administration and consulting at no cost, allowing faculty contributors to stretch their research funding. In exchange, the BRC Program can provide unused capacity on the contributed compute servers to other campus researchers, increasing the reach and impact of research investments.

Savio Expanding

To accommodate additional users, Savio’s parallel file system was doubled in early 2015, and a further upgrade to over a petabyte of fast parallel storage will come online in summer of 2015. In addition, the BRC-HPC team recently worked with the IST Networking team to connect Savio’s Data Transfer Node (DTN) to the campus' new Science DMZ 100GBE network, enabling fast transfer of large data resources located beyond UC Berkeley. These features are intended to facilitate “Big Data” research requirements for streamlined data transfer and fast, high-volume storage.

Savio is currently expanding, and the initial campus contribution of compute nodes will more than double by this summer. The cluster will get another 72 standard nodes, 12 nodes specialized for data-intensive “multi-task computing”, and 6 nodes with GPU (Graphics Processing Unit) accelerators, bringing the total Savio capacity to nearly 5500 processor cores (plus nearly 70,000 GPU cores).

Faculty Computing Allowance

The BRC Program is launching a service this spring that offers a free allocation of computational resources to every faculty member at Cal with a project that can benefit from Savio. This will initially be available as HPC resources on Savio, but in the near future will support allocations on a more diverse range of compute needs and models. The goal for future allocations is to include cloud-like computation models, as well as "virtual workstations" equipped with greater computing power, storage, data transfer speed, and/or memory than is typical of a laptop or desktop workstation.

 

These categories of computational research resources, and the BRC community of consultants that support their use, are described more fully on the BRC page of Research IT’s web site.

For inquiries about access to BRC computational resources, consulting, and/or server contributions to Savio, please contact Research IT at research-it@berkeley.edu.

 

Tags: 

Program: 

Service: 

Partnership: