Behind the scenes at NASA’s computer powerhouse

Whether the mission is to discover new planets or study the one we call home, the scientists at the NASA Ames Research Center are constantly relying on one priceless member of the team. Her name is Pleiades; she doesn’t talk much.

Pleiades, a vast 210,000-processor supercomputer housed in a rear office building at the NASA Ames campus, might be considered the central nervous system for government research. Twenty-four hours a day, seven days a week, Pleiades is churning away at hundreds of unique simulations and algorithms shared between government scientists coast to coast. Among its common tasks, the supercomputer is analyzing Earth’s ocean currents and the durability of a space probe’s fuselage, and determining if a distant glimmer of starlight may reveal an undiscovered planet.

The power of this system is hard for a lay person to fathom. In the parlance of computer engineers, Pleiades is measured to consistently run at a processing speed of 4.09 petaflops, which basically refers to the number of operations the machine can do in one second.

Piyush Mehrotra, chief of the NASA Advanced Supercomputing Facility, likes to use a different analogy to demonstrate that speed.

“If you had every human on Earth doing calculations, it would take 300 years to match the calculations Pleiades can do in a second,” he said, matter-of-factly. “Think about it as 40,000 to 50,000 workstations or laptops, the kind that you’d have at home, working together simultaneously.”

Waving his key card in front of a door scanner, Mehrotra opened the door to a cavernous room containing row after row of Pleiades’ monolithic computer towers. Each tower stood about 6 feet tall and contained stacks of processor nodes. Like any data center, the Pleiades computer room didn’t appear as some dazzling spectacle of information coursing at the speed of light. But the unseen power was still palpable — the room echoed with a throbbing hum and the lanes between computer rows felt like wind tunnels from the blaring cooling fans.

Surprisingly, the engineering of Pleiades isn’t much different from a home personal computer. At one time, supercomputers were almost entirely custom-built machines filled with a handful of specially designed chips streamlined for performance.

In the mid-1990s, NASA computer engineers were among the first to break out of that mold by building a supercomputer based on commodity parts — that is, the same chips available on the retail market. Clustering a multitude of cheap mass-produced processors turned out to be a much cheaper way to maximize performance.

Today, the 210,000 processors that make up Pleiades are all various generations of the Intel Xeon chipset. Each year, the Advanced Supercomputing Division undergoes about $15 million in upgrades, during which the oldest generation of processors are phased out for the newer replacements.

Does NASA get a good deal for buying in bulk?

“Well, we should certainly hope so,” Mehrotra said, smiling. “We tend to get good prices because we’re buying big huge systems and upgrading all the time.”

Following its most recent round of upgrades, Pleiades was pegged by one measure as the third-fastest supercomputer in the U.S., trailing only the federal Department of Energy’s two massive computer centers.

Yet there are times when even one of the world’s fastest computers can still seem too slow. There is a near-insatiable demand for processing power between NASA’s four separate divisions as well as the universities and outside groups wanting the supercomputer for government research.

NASA directors dictate which projects get priority on Pleiades, and it falls down to Mehrotra and his team to figure out how to allocate the machine’s brainpower accordingly. Naturally, they have algorithms to help portion out 300 to 400 research projects at any given time.

Among the foremost research projects being run on Pleiades is an oceanic study with a long-winded name: Estimating the Climate and Circulation of the Ocean 2 (ECCO2). This study, started in 2011 as a partnership between several federal research agencies, seeks to create a worldwide ocean model to track sea currents, heat patterns, sea ice, salinity and other factors.

The simulation is performed practically in real-time by collecting data from various ocean instruments and space satellites. Pleiades is taking this data from a patchwork of sources and collating it into one global model, explained Christopher Hill, ECCO’s principal research engineer.

“We designed this study to be able to take advantage of computers as they got bigger,” Hill said. “On a system like Pleiades, we can do thousands of times more than we could on a laptop computer.”

ECCO2 has been the most demanding study performed to date on Pleiades, at times taking up more than one-third of the computer’s processing power and producing more than three petabytes (three million gigabytes) of raw data. That information has been used in hundreds of published studies and research papers.

The Ames supercomputer is also heavily involved in more traditional roles of NASA space exploration. Pleiades is routinely used to perform simulations on materials used for spacecrafts. Perhaps its biggest success to date, the supercomputer played a lead role in sifting through reams of data taken by the Kepler spacecraft to discover about 1,000 confirmed planets and about 4,200 more candidate planets.

But the evolving capability of Pleiades also makes it a high-value target for hackers. Every day, Mehrotra said, his system is being bombarded by unauthorized users from across the world trying to gain access.

If Pleiades’ firewall were breached, a hacker could try to use the supercomputer for malicious attacks on other computer systems. So far, no such attempt to break into Pleiades has been successful, Mehrotra said.

Mountain View plans substantial road repaving starting April 1520 hours ago

One dead, two injured in Los Altos Hills I-280 crash22 hours ago

With new plan, Palo Alto prepares to go big on housing around San AntonioApril 6, 2026 11:20 pm