AsianScientist (Jan. 25, 2018) – When dealing with the vast distances involved in interplanetary travel, Earth-based navigation systems can be off by hundreds of kilometers. But what if spacecraft could triangulate their own positions using signals from pulsars—neutron stars that emit radiation at highly regular intervals—just as sailors used to navigate by the heavens?
This is more than just conjecture—a pulsar-based navigation system is already being tested aboard the International Space Station (ISS). But this and other applications for deep space exploration are likely to run up against a major obstacle: the lack of compute power in space.
“If you travel to the moon, you can still rely on Earth-based computers to provide answers within seconds. But on Mars, it could take 20 minutes to transmit a message and another 20 minutes to receive an answer,” Dr. Goh Eng Lim, vice president and chief technology officer, high performance computing (HPC) and artificial intelligence (AI), Hewlett Packard Enterprise (HPE), tells Supercomputing Asia. “As you travel further from Earth, you’ll need to carry more computing resources with you.”
Goh is the lead investigator on a joint HPE-NASA endeavor to build a supercomputer which can weather the harsh conditions of space. In September 2017, the Spaceborne Computer—a two-node machine weighing about 60 kilograms on Earth—was installed on the ISS, where it will undergo a year of rigorous testing.
The app store at the end of the universe
In space, cosmic radiation wreaks havoc with computer circuitry, causing frequent glitches. Computers thus need to be physically ‘hardened’ or shielded against radiation. But by the end of this expensive, time-consuming process, they may be several generations behind the latest models.
Instead of hardware modifications, the Spaceborne Computer uses software to slow down its operations in order to prevent damage during a radiation event, meaning that it can be used straight out of the box.
Computers on space missions today also tend to have very specialized functions. “You have to think very carefully beforehand about what precious applications you intend to carry with you, because you will be embedding them in those hardened computers,” says Goh. “If something new comes up, it’s more difficult for them to run applications that are unanticipated.”
But one-trick ponies are not ideal for the vagaries of a round trip to Mars. For longer, more complex space missions, having general purpose capabilities will minimize the need for astronauts to improvise in the event of an emergency.
The Spaceborne Computer, with its off-the-shelf hardware and Linux-based operating system, has the potential to be this all-rounder. Goh envisions space missions using it the way we use smartphones on Earth—for everything.
“Before launch, you can sit down and load all the applications you think you might need on this commercial off-the-shelf (COTS)-based computer. As you fly, if something unanticipated comes up, you hope you have preloaded the relevant app somewhere on the system,” he explains. “This is a very powerful concept.”
On the ISS, the Spaceborne Computer has achieved speeds of up to one teraFLOP—an order of magnitude more powerful than anything else in space, though still a far cry from its much larger Earth-bound counterparts.
But even if we could launch the world’s fastest supercomputer into space, its power needs would quickly suck the spacecraft dry, says Goh. Traditional high-performance computing uses physics-based models to run simulations and make predictions—a top-down, computationally intensive approach.
Launching more compute power into orbit is thus not the only goal. What astronauts also need is a lightweight, versatile computing system that can help them make quick decisions in a pinch.
This is where machine learning comes into the picture, says Goh. Here, instead of traditional physics-based models, computer algorithms would learn by ingesting large amounts of data from past simulations, and then assign weights to a range of parameters according to their importance. These weights can then be used to make inferences, or predictions, about a given situation.
“This is what you carry with you when you travel—just the weights—and you can do lightweight predictions because you’ve done all the heavy lifting beforehand [on Earth],” Goh says. “I believe this is the approach we will quite often need for long duration space travel.”
Scaling Asian supercomputing
Today, increasingly massive amounts of data are available for machine learning algorithms to ingest, meaning that they will have to do so more quickly. But many machine learning approaches, in particular deep neural networks, do not scale well, said Goh.
Back on Earth, the Tokyo Institute of Technology (TiTech) is working to address this very problem. “TiTech came to us with a need—to scale artificial intelligence (AI). It was a very clear, simple mandate,” says Goh.
HPE built TiTech the Tsubame 3.0 supercomputer, a 47-petaFLOPS “massive bandwidth machine,” as Goh calls it, with multiple tightly connected nodes that voraciously consume and crunch data. One of the fastest AI supercomputers in the world, its architecture also lays the groundwork for machine learning to be used in concert with more traditional high-performance computing approaches.
“Physics, of course, will need to be there as it always has, but you need to complement it with machine learning,” says Goh. “Increasingly, we are seeing physics-based, top-down methods of prediction working side-by-side with bottom-up machine learning algorithms.”
When it comes to pioneering new approaches in computer science such as this, Asia can rely on one advantage—its large number of STEM graduates, thinks Goh.
“Smart people are everywhere, but to solve a problem you need a lot of them to come together. In Asia, there’s still a very strong attraction for people to study STEM, and Confucian systems tend to give more reverence to education.”
Still, whether Asian governments have the appetite to support bigger, more profound projects with no immediate economic returns remains an open question. Take the Laser Interferometer Gravitational-Wave Observatory (LIGO) in the US, for example—its Nobel prize-winning detection of gravitational waves was decades (and hundreds of millions of dollars) in the making.
“If the gravitational wave pitch were given worldwide, would any Asian government take that up? I think that’s a good question to ask. The [US] National Science Foundation did, so credit to them—they also have their funding challenges, but it was profound enough for them to take it up,” says Goh.
Advances on the cards
In the future, supercomputers in space could benefit from an earthly pursuit—poker. In Go and chess, players are privy to their opponent’s position on the board. But poker is a different beast—not only do players not know their opponent’s cards; they also go out of their way to mislead one another.
Incomplete information is often part and parcel of real-world situations—consider contract negotiations, auctions and military strategy, for example. The same is true for space travel.
“On Earth, you can amass resources to get information to make decisions,” says Goh. “But in space, you have less time and resources to do this.”
In January 2017, a poker-playing supercomputer built by HPE and researchers at Carnegie Mellon University in the US handily beat four top professional players. Its specialty—out-bluffing humans—represents a milestone in AI.
“The goal is to translate this into an AI system that can handle and make decisions based on incomplete information,” said Goh. “This is very relevant because you need to be more self-sufficient the further out you go from Earth.”
While the harsh conditions of space can complicate even the simplest of tasks, there is one thing that is easier than on Earth—keeping supercomputers cool. The Spaceborne Computer is chilled with a fluid-cooling system that simply circulates through the frigid shadow side of the ISS.
“One could say it’s the greenest supercomputer on—I almost said on Earth and I would have been wrong,” laughs Goh, who adds that he is clearly on new ground. “One could say it’s the greenest supercomputer built by humankind.”
This article was first published in the print version of Supercomputing Asia, January 2018. Click here to subscribe to Supercomputing Asia in print.
Copyright: Asian Scientist Magazine.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.