Power-efficient datacenters are practically a fact of life at this point: it seems as if every tech giant from Google to Facebook is trumpeting their respective datacenters’ reliance on renewable-energy sources.
Supercomputers aren’t exempt from this focus on “Green IT.” Just last week, for example, the NCAR-Wyoming Supercomputing Center (NWSC) won the ‘Green’ Data Center of the Year award from datacenter Website Datacenter Dynamics.
The site, which also won the Uptime Institute’s Green Enterprise IT Award, made its mark with the design for a 153,000-square-foot facility built in two modules, each of which holds its own heating, cooling, power and ventilation.
The facility, in Cheyenne, Wyo., was sponsored by the National Science Foundation for atmospheric research and was designed to be flexible enough to adapt to changing environmental conditions and computing demands. The modular design of the NWSC building means the facility can add 50 percent more space by dropping in another module. The internal design is modular as well, allowing for quick and comparatively cheap resizing of power and cooling components.
The supercomputer itself is an IBM iDataPlex with 72,288 processor cores, 144.6 terabytes of memory and a Mellanox FDR Infiniband interconnect to tie things together. Its top speed is 1.5 petaflops, which puts it at 17 on the list of Top500 fastest supercomputers. Outside air provides the majority of the cooling, and heat from the supercomputer is recycled to warm the building and melt snow on walkways. Ten percent of total power the facility uses comes from wind power; that number will grow over time as more wind-generation facilities are built.
All told, the building and support systems use less than 10 percent of the power consumed by the whole installation; more than 70 percent of the building is made with concrete, wood, metal and other materials diverted from landfills or recycling centers.
That isn’t the only super-energy-efficient supercomputer in development. Researchers at MIT’s Lincoln Laboratories have spent more than a decade working on a system that’s both powerful and “green.” The resulting TX-2500 platform features 1,500 processors, a petabyte of disk space spread across 400 servers, and a 6-disk 1.5-terabyte RAID storage system connected to each node.
The system went live in 2008; in 2009 LLGrid ran what was at the time the largest single calculation ever performed on a single computer: a one-half-petabyte calculation for the HPC Challenge STREAM benchmark.
TX-2500 was built as part of the Labs’ larger LLGrid system, which includes the Massachusetts Green High Performance Computing Center (MGHPCC). Located 90 miles from Lincoln Labs in Holyoke, Mass., that facility is built inside a former textile plant on the Connecticut River, a cheap source of hydroelectric power. Seventy percent of the power for the site comes from renewable sources, including hydroelectric and solar.
Like the Wyoming facility, MGHPCC was built in modules—two Hewlett-Packard EcoPODS, or container-like boxes filled with datacenter gear that can be plugged together to add or subtract capacity more easily than building from the ground up. Each EcoPOD has its own cooling system and space for 44 racks that can hold 24,000 hard drives. The center as a whole can house 1,500 computer nodes, .4 petabytes of memory running at a top speed of half a petaflop. It is connected to its users via 10Gbit/sec fiberoptic cables.