The ISS Has a Supercomputer Never Mind the Fried Disks
wired.com
wired.com
“Supercomputers” are now only 30-100 times faster than a mobile phone?
Not sure if that says more about the “supercomputer” or the phone.
This was also amusing:
The computer also remained operational even thought it lost its primary upload and download connection to NASA up to eight times a day for periods ranging from 3 seconds to 20 minutes. That kind of independence might prove useful on an eventual moon base or mission to Mars, to work around communications delays back to Earth.
As if a computer that continues running when its communication link goes down is an accomplishment?
Probably because "kids these days" don't actually know any computers exist, that aren't mobile phones (or phone OS based tablets). Its a stupid comparison, probably added to make the article catchier.
SuperComputer in this case is more akin to a "really powerful desktop"
30x from iPhoneX to really powerful desktop seems reasonable?
Aka a "computer".
These days? Compared to the typical case? Yes!
There was the story of Chrome (IIRC) crashing when it lost internet access. (Yes, it should be unable to access websites, but it shouldn't crash.)
JIRA will stutter on text input if its AJAX API calls fail or are too heavy.
I've seen developers set up a local mode of an app that somehow sneaks in an internet dependency.
Not even going to review Electron.
Similar overcoupling in the recent story about Google "no longer supporting login" if you're not running JavaScript.
So yeah, it shouldn't merit praise, but a lot of systems fail even at that.
It shouldn't be, but it seems like it is.
My Apple machines, in particular, can't stand a slow net connection.
And I'm not talking just about third-party apps. For some reason Macs are good with either huge bandwidth, or zero connectivity. Anywhere in between is problematic. This goes back to at least the 1990's, when Airports had dialup modems in them.
The only thing novel about this is that they sent a computer designed in this millennium and it survived the radiation unshielded.
Regarding the SSD/NAND flash, it makes sense that it would die relatively quickly. It's constantly being bombarded by energetic particles and the cell sizes, geometries, and write voltages are low and small enough that I could see bit flips accumulating relatively quickly from cosmic rays. The drive's firmware would then determine that the write block had failed and mark it permanently as bad. What's probably needed here is a different firmware that attempts to erase the block and then tests it and puts it back into service if it passes. Or different cell geometries and higher write/erase voltages are needed. It's an interesting note.
What this experiment is about is satellites. Which is a big deal. If you can put stuff in LEO and use COTS parts, that saves a ton of money. And a super computer is much bigger than the testing they've done with small systems and cube SATs. Moon hardware will always have to be rad hard, but this shows that the commercialization space can be cheaper than was previously thought.
Earth's ground is created by the giant revolving metal core. the Moon doesn't have that. So how do you do that for big permanent systems on the Moon, or Mars for that matter?
Rather I should say "How do you do it safely?", because it definitely is possible already.
That doesn't sound that super to me! "100 times faster than a phone" just doesn't have an impressive ring to it.
https://arstechnica.com/science/2017/08/spacex-is-launching-...
It's sort of funny how many of the flaws of old computing hardware turned out to be hugely beneficial for use in space. Bulky transistors that are relatively safe from bit flipping, high power draw that means small fluctuations don't cause outages, limited processing and storage that forced efficiency and lowered the amount of stuff exposed to radiation.
I'm not sure where the tipping points would be, but it wouldn't surprise me if getting a 386 working in space is easier than getting a Pentium 4 to achieve the same things.
The Spaceborne Computer includes the HPE Apollo 40 class systems with a high speed HPC interconnect running an open-source Linux operating system. Though there are no hardware modifications to these components, we created a unique water-cooled enclosure for the hardware and developed purpose-built system software to address the environmental constraints and reliability requirements of supercomputing in space. Generally, in order for NASA to approve computers for space, the equipment needs to be “ruggedized” – or hardened to withstand the conditions in space.
https://news.hpe.com/hewlett-packard-enterprise-sends-superc...
For that matter, it doesn't look like stage one of putting a supercomputer into space. (And really, why would you want that?) From that last paragraph, its tasks from now on will be running sample takeoff and landing calculations, which implies this is about testing whether modern tech can replace existing flight computers.
Basically before they bought Compaq and sold off Agilent and when they produced calculators that weren't made of poop.
The times change. Back when I worked in the MCC building, newcomers with an HP calculator were told "you better not bring that into the building again."
The chief flight director (Gene Kranz) apparently had a thing about HP.
This is good on the basics: https://space.stackexchange.com/questions/3702/does-the-iss-...
The SpaceX system was quoted with pretty good latency numbers, but has the added complexity of multiple satellites and everything moving. I've grown used to knowing which servers are nearby which other servers.
From a practical side, remote hands is going to be expensive, and who are you going to impress with all the fancy security in your lobby etc.
Unfortunately, there are a few orders of magnitude difference between the ISS and a data centre. The ISS solar panels produce up to 120 kilowatts [0], assuming the station is in the sun. A large data centre can use power upwards of 30 GWh [1].
Launching nuclear reactors is also somewhat controversial.
[0] https://en.wikipedia.org/wiki/Electrical_system_of_the_Inter...
[1] https://www.techuk.org/images/programmes/DataCentres/Data_Ce...
It looks like the biggest reason (other than "it's 20+ year old tech") is that triple-junction panels weren't used to save weight. Rather than optimizing power-per-inch or power-per-manufacturing-cost like we would on Earth, NASA was targeting power-per-ounce, which meant gathering only the highest-power wavelengths. Add to that some issues with panel degradation from radiation, and it's a surprisingly tough way to gather energy long-term.
Short of sci-fi stuff like a space elevator or extraplanetary manufacturing, I'm not sure how solvable this is. Does anyone know of ways of getting high power output in space other than nuclear?
The only thing we have at the moment is chemical rocket fuel. This is obviously impractical for an orbiting data centre because of the continuous, high launch costs.
The ISS cools itself with radiators using ammonia.
https://www.nasa.gov/content/cooling-system-keeps-space-stat...
https://en.wikipedia.org/wiki/External_Active_Thermal_Contro...
The mind boggles trying to imagine what a data-center sized cooling system designed to reject heat solely through radiation and optimized principally for weight looks like.
Shielding will mitigate a lot, but redundant hardware, error detection and correction, and resilient software will also be needed. It's a vastly more hostile environment for digital devices than anywhere on earth.
Astronauts are a pretty small, fit population that get a lifetime of top-notch healthcare and monitoring, so it's a little hard to isolate exactly what their risks are.
That said, airline pilots are probably at greater risk, given the decades worth of exposure they receive.