12 ms·
The ISS Has a Supercomputer Never Mind the Fried Disks
- setquk 8y agoThis feels like "old HP" aka "good HP" back when they were at the forefront of innovation. Basically before they bought Compaq and sold off Agilent and when they produced calculators that weren't made of poop.
- roywiggins 8y ago> The experiment was designed to test the performance of the HPE Spaceborne Computer, which is comprised of 32 separate cores that work together and can run operations 30 to 100 times faster than a standard iPhone or tablet That doesn't sound that super to me! "100 times faster than a phone" just doesn't have an impressive ring to it.
- Bluecobra 8y agoIt is compared to the 386 processors on the main command computers. :) https://arstechnica.com/science/2017/08/spacex-is-launching-a-supercomputer-to-the-international-space-station/ https://arstechnica.com/science/2017/08/spacex-is-launching-...
- ljcn 8y agoCool. The control machine seems like a bit of a waste though. Hardware degrades randomly even on Earth, it'll be hard to make any meaningful comparison with a single sample.
- Bartweiss 8y agoAnd it looks like it's not a clear winner over those, either. 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.
- ReverseCold 8y ago32 cores makes it sound like they bought a regular Xeon or EPYC chip and sent it to the ISS
- ams6110 8y agoPretty much. 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-supercomputer-into-space-to-accelerate-mission-to-mars/ https://news.hpe.com/hewlett-packard-enterprise-sends-superc...
- Bartweiss 8y agoYeah, not sure where the article came up with that word. It looks like this was about testing a custom variant of a server/supercomputer processor to see how long it would hold up, yeah. So presumably ECC and the 32 cores are on one or two processors, but definitely not an actual supercomputer or even the high end of single chips. 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.
- znpy 8y agoI do not expect much more from Wired.
- ape4 8y agoIf half the disks are fired, how are the astronauts!
- ceejayoz 8y agoHumans have much better self-repair mechanisms. 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.
- sebazzz 8y agoIs it possible to better shield them or would that require putting the computer in an airtight lead box?
- adrianN 8y agoCosmic radiation has a lot of energy so the lead would have to be fairly thick. Getting heavy things into space is expensive. A kilogram costs on the order of $10k.
- scrumper 8y agoThe article discusses that a bit. Water shielding is quite effective. Doesn't have to be airtight. 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.
- fjsolwmv 8y agoCan mods fix the grammatical abomination that is the title?
- joshstrange 8y agoNow I'm just thinking of the logistics of running a datacenter in space. Energy and cooling seem like easily solved problems (solar and, well, space as solutions respectively) but I'm sure upload/download isn't great. That said SpaceX's satellite internet stuff was promising gigabit speeds IIRC. And obviously the shielding would be an issue.
- scrumper 8y agoCooling's not that easy in space, quite the opposite. You're limited to radiative cooling. Cooling the ISS itself is quite a challenge. This is good on the basics: https://space.stackexchange.com/questions/3702/does-the-iss-need-more-heating-or-more-cooling https://space.stackexchange.com/questions/3702/does-the-iss-...
- toast0 8y agoUpload and download capacity shouldn't be too bad. SpaceX was promising gigabit to multiple customers. Latency will be an issue though --- assuming you want geostationary for stability of routing and ease of aiming dishes. 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.
- plopz 8y agoWikipedia says SpaceX is planning on two groups of satellites at 340km and 1200km, so it would be low Earth orbit rather than geostationary, which is how you solve the latency problem.
- toast0 8y agoYeah, and that's fine for comms satellites. It's not so great for a 'datacenter in space' because now your data is variable latency.
- 8y ago
- jschwartzi 8y agoIt doesn't sound like they're doing anything novel compared to computing in harsh environments here on Earth. It's pretty routine to monitor the output of computation and cross-verify when an error is hazardous and there is risk of corruption. Frequently when the system is used for some critical task and there's an error with a less critical subsystem it will operate in a "degraded state." And ECC RAM has been standard on these kinds of computers for ages. 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.
- hindsightbias 8y ago> Hewlett Packard 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.
- rjsw 8y agoI thought that the Space Shuttle used an HP 41c for some extra navigation functions, maybe that was after Gene Kranz had retired.
- PascLeRasc 8y agoWeird, older engineers I know all love HP calculators for the RPN.
- tlrobinson 8y agocan run operations 30 to 100 times faster than a standard iPhone or tablet “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?
- octorian 8y ago> can run operations 30 to 100 times faster than a standard iPhone or tablet 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.
- sp332 8y agoThe iPhone X, which is the fastest phone I can think of, gets around 500-600 gigaflops which would put it just off the Top 500 list in 2004. If we multiply by 30 for 15 Tflop/s, that's the bottom of the Top500 in 2009. Last year the bottom was 400-500 Tflop/s, which is about 1,000 iPhone X's. I mean you don't have to be in the top 500 to be a supercomputer, but I would expect it to be several hundred and not around 30.
- brianwawok 8y agoSuperComputer from the Top500 list is a rack if not a room full of many of thousands of pounds of computers. SuperComputer in this case is more akin to a "really powerful desktop" 30x from iPhoneX to really powerful desktop seems reasonable?
- romwell 8y ago>"really powerful desktop" Aka a "computer".
- godelski 8y agoIt sounds like the author missed the purpose of the experiment, when they mentioned the moon. This would not work on the moon. The ISS is within the radiation belts and has a lot of protection. 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.
- Sir_Cmpwn 8y agoBury it a few feet and it'll work fine on the moon. With off-the-shelf rad-unhardened tech, too.
- TeMPOraL 8y agoThat works for stuff you want to put on the Moon permanently. But the ships that go there will need computers too, and so will surface installations.
- godelski 8y agoWhich brings up a fun thing that few people talk about: How do you ground? 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.
- godelski 8y agoThat _should_ work. But needs to get to TRL 8/9. Which was the point of this experiment. To boost the TRL.