ACADEMIA

Whether you’re off to the ballpark, a picnic or a round of golf, accurate weather forecasting is crucial to your preparation. The Weather Channel is working to improve weather predictions in places just like these using a high powered supercomputer to run their innovative weather prediction algorithm called High Resolution Aggregated Data or HiRAD. To accurately predict weather conditions, The Weather Channel combines a plurality of current meteorological data. They start with about 1,600 established observation points like airports, military bases, etc, and then combine this with weather radar data, and other resources. Using an HP ProLiant DL385 cluster, this data is calculated with The Weather Channel’s patented HiRAD system to infer the weather at 1.9 million points of interest in the US. HiRAD will predict weather conditions such as wind speed, precipitation, and temperature within two degrees of the actual temperature. “The unique thing is that we’re assimilating a collection of live weather data that’s normally used separate and apart from each other, to create a single integrated and nearly continuous field of information,” Dorren Schmitt, Senior Unix administrator for The Weather Channel, said. “Rather than guessing your local weather based on reports from observation points typically found only at nearby airports or military bases, you have accurate information for specific points of interest like Dodger Stadium.” In January 2005 when The Weather Channel began applying the HiRAD system, they were limited to 10,000 calculated observation points. Schmitt and other team members took the challenge of increasing the calculated data points from 10,000 to 1.9 million and began building a system support architecture that would support it. Just 11 months later they went live with the added data points, using HP hardware and Cluster Resources’ Moab Cluster Suite to run their memory-intense application. Running the calculations takes 97 percent of the clusters’ CPU and 98 percent of the RAM. In order to constantly run the complex calculations without using up the valuable time of the administrators and developers, The Weather Channel employs Moab’s event policy engine — a tool used to automate administrative tasks — to monitor and command the system to run calculations three times per hour, 365 days per year. “Moab has been rock solid since the beginning,” Schmitt said. “It kicks off beautifully like clockwork needing no human intervention, which is unlike so many other schedulers. There weren’t very many environments out there that allowed us to do this level of automation where we set it up and it drives itself.” Moab allows The Weather Channel to specify exactly when they want a calculation to run, how often, and on which hardware. Such fine-grained control lets administrators extract the maximum value out of their clusters. “We designed Moab to take the burden off administrators,” Dave Jackson, CTO of Cluster Resources said. “Moab’s event policy engine empowers administrators by letting them set policies that define exactly what they want to have happen. Moab automatically enforces these policies, letting administrators put their clusters in auto pilot and freeing them to work on more pressing projects.” The Weather Channel is using a HP ProLiant DL385 cluster using two Dual Core AMD Opteron CPUs/per node with 16 gigabytes of RAM per CPU. The system runs Red Hat Enterprise Edition 4.0 as an operating system, Veritas Foundation Suite for file systems, PBS Pro for basic queuing, and Moab Cluster Suite 4.2 for scheduling and workload management. “Cluster computing is providing supercomputing performance to organizations that want access to considerable computing power but don't have a big IT budget,” said Bruce Toal, director of marketing for High Performance Computing Division at HP. "HP is working with partners like Cluster Resources to provide solutions that allow customers to tackle larger, more difficult problems than ever before.” To try Moab’s event policy engine and other features free for 30 days, visit its Web site or call (801) 717-3700.