Following a February launch, HPE’s second Spaceborne Computer (SBC-2) has been circling Earth on the International Space Station for some six months. The first Spaceborne Computer had returned to Earth around 20 months prior – and even before then, HPE had been hard at work distinguishing the sequel from its predecessor.
“In retrospect, Spaceborne-1 is a proof of concept,” said Mark Fernandez, HPE’s principal investigator for SBC-2, in an interview with HPCwire. NASA had, he explained, initially tasked HPE with three major questions for spaceborne computing: first, can you take off-the-shelf components and put them into space? Second, can they survive the launch and be installed by non-IT experts? Third, can they actually function in space – and if so, for how long?
That mission – optimistically planned for a year – instead lasted 1.8 years thanks to smart planning and software-driven “hardening” that slowed the system’s operations when parameters indicated that dangerous space weather might impact the hardware. In that time, Fernandez said, the system ran over 50,000 software experiments, with every last one returning the correct result.
Understandably, this had NASA excited about a successor. “Before we returned to Earth, NASA asked us if we could do it again,” Fernandez said. “But they had some major changes.”
Planning a second sojourn
Three big bullet points distinguished the SBC-2 mission from the SBC-1 mission. This time, HPE had no requirements for the system, instead solely using the workloads to serve the space and Earth observation communities. NASA, for its part, asked that HPE send up twice the hardware for redundancy – and, ambitiously, that the mission last two to three years: the estimated length of the first missions to and from Mars.
In the end, they sent up a matching set of HPE Edgeline Converged EL4000 Edge and HPE ProLiant systems, each including an AI-focused node with an Nvidia T4 GPU. “We’re now sending up more than twice the cores, and they’re faster,” Fernandez said, explaining that the new hardware delivered more than two Linpack teraflops – over twice that of its predecessor. “All the hardware is stock,” he added. “All the software is Red Hat 7.8, unmodified.”
Emboldened by the success of SBC-1, HPE was also more comfortable changing up the software on the systems. “This is unlike previous space missions where things are locked and loaded prior to launch and ‘thou shalt not change it,’” Fernandez said. And keeping the software stock, he said, meant that developers on the ground could easily workshop software for use on SBC-2.
On the edge of space
All of this software flexibility opened up new possibilities for how the system could handle computing at one of humanity’s most extreme edges. “During Spaceborne-1, the most often-asked question was, ‘if you could just gzip up my data, that would really, really help!’” Fernandez said. During SBC-1’s tenure, they would do just that – but when the same requests started popping up for SBC-2, the situation was different.
By way of example, Fernandez mentioned a partner (“who shall remain nameless”) that wanted to work with a 180GB dataset generated on the space station. Gzip got the size down to around 18GB – a respectable 90 percent compression. “They were super, super excited. Instead of taking 12 hours to download, they said, ‘I might be able to get this in an hour or so!’” Fernandez recounted.
But then, he asked what they planned on doing with the dataset. Upon learning that they just wanted to run an industry-standard software package on it, he asked the partner: “Well, do you want me to run that for you?”
“And that’s when the lightbulb went off,” he said. After just a few minutes of CPU/GPU processing, he explained, the work was done. “Without touching their code, I ran it, and we ended up with 20,000x reduction … and I’m able to download it now in two seconds.”
That sort of experience hasn’t been an isolated incident with the advent of SBC-2. Fernandez said that Microsoft, as well, had partnered with HPE to process human genome data at the edge, resulting in a 10,000-fold reduction in the size of the download.
Enabling space-based research
SBC-2 is hosting dozens of experiments, with some already completed. Many of these, Fernandez explained, related to operations within the ISS, such as health-related studies of the astronauts’ vitals and genomes, life sciences work on the plants aboard the station and preliminary work on autonomous mission operations. (Fernandez said that this latter area of research is targeted at longer missions, with NASA aiming to ameliorate the long-term psychological impacts of rote tasks on spacecraft.)
Outside of the walls of the ISS, SBC-2 experiments are honing in on a range of feature extraction tasks. Researchers, Fernandez said, are aiming to avoid streaming data-intensive ultra-HD video in its entirety by processing the video on SBC-2 and identifying events like wildfires, lightning strikes and illegal fishing vessels before transferring the results to Earth.
Other use cases include work on satellite algorithms and encryption and an educational outreach program that allows students to compete to have their code run on SBC-2. Fernandez also mentioned the possibility of using SBC-2 to test the fundamentals of swarm-based satellite processing, which would allow the use of lighter-weight, lower-power satellites that all communicated with a spaceborne computer.
Looking to the stars
“I was briefing some people this morning about Spaceborne-3,” Fernandez said. “And it may be 3A, 3B, 3C, because there [are] multiple desires for the functionality for Spaceborne-3.” These desires, he said, included the use of spaceborne computers to help upgrade the station’s antiquated IT infrastructure and working to expand the systems’ storage capacity and capabilities.
Furthermore, work remains to be done in the automation and fine-tuning of the spaceborne computers’ software-based hardening against the ravages of space. HPE’s eventual aim remains to apply machine learning to this task, but, Fernandez said, data collection is still ongoing. Whether that functionality falls within the tenure of SBC-2 or SBC-3, he said, depended entirely on how many anomalies SBC-2 encountered. “We’ve had a few,” he said, “but not enough yet to have a confident, well-trained model.”
Broadly, HPE is continuing to pitch the spaceborne computers as an extreme case study of a basic proposition. “If you can do it in space, you can do it anywhere,” Fernandez quipped. “We can compute faster than we can download. This applies at the edge wherever you are – whether you’re on an oil rig, or you’re on an aircraft, or you’re in a cell phone tower.”