Scaling (UP) Olympus
PSC Public Health Applications Group Leverages Bridges Construction to Upgrade Modeling Cluster
PSC’s work with Hewlett Packard Enterprise (HPE) and Intel on its new Bridges system has paid dividends for the public health community in the U.S. and worldwide.
Leveraging the ongoing relationship with these vendors, PSC’s Public Health Applications Group is upgrading Olympus, the dedicated computer cluster it administers for the Models of Infectious Disease Agent Study (MIDAS) Research Network. The MIDAS network is a National Institutes of Health-funded international consortium of infectious disease transmission modelers.
“The upgrade to Olympus will provide computing capability to MIDAS and greatly increase the performance and capacity of the cluster,” says PSC’s Shawn Brown, leader of the MIDAS MISSION Group, which organizes MIDAS software developers around the world to share software, provide best practices in development and validation and gather requirements for userfocused tools. “The system is particularly designed to provide state-of-the-art capability for performing public health simulations of infectious disease spread and how to stop them.”
The upgrade will add 40 HPE Apollo 2000 nodes, half with 128 gigabytes (GB) of shared random access memory (RAM) and half with 256 GB of RAM. An additional node, HP DL580, will have a full 1 TB (1 terabyte, or 1,000 GB) of RAM. By comparison, many standard supercomputers have 128 GB of RAM per node. The improved Olympus will improve on its predecessor with 1.5 times the computational speed, 2 times the data storage, 1,000 times the network speed and advanced Big Data capabilities stemming from addition of the new 1 TB category of nodes.
“Together, these components will provide a transformative ‘Big Data’ capability that’s unique among dedicated public health computational systems,” Brown says.
PSC LOOKBACK: CAPTURING THE TWISTER
While PSC’s Public Health Applications Group is a relatively new effort, it isn’t the center’s first foray into improving public safety. In 2004, Ming Xue of the University of Oklahoma used PSC’s Hewlett Packard-delivered LeMieux supercomputer to create the first realistic simulation of tornado formation. In a period when tornado warnings were three times as likely to be false alarms as lifesaving alerts, PSC’s terascale system allowed Xue to simulate a “supercell” thunderstorm at a scale of 20 to 25 meters. That was far better than the 100-meter resolution of previous simulations and the first that simulated at a finer level than the typical tornado path width of 46 meters. Just like the real 1977 storm it was modeling, the LeMieux simulation formed a funnel cloud, paving the way to today’s lifesaving tornado warnings.