Engineers building first space supercomputer

PhysOrg.com ^ | October 26, 2006 | University of Florida

[annie laurie]

Ping

[Paleo Conservative]

If you want on or off my aerospace ping list, please contact me by Freep mail.

[UnbelievingScumOnTheOtherSide]

100 times faster than any computer in space today

100 times faster than the TRS-80s on the shuttle.

[Jedi Master Pikachu]

From a generally unknowledgeable viewpoint, it seems trying to circumvent destruction to a computer via adaptable software, is not such a smart decision. What if all the boards get destroyed by radiation? Is it only temporary?

[free_at_jsl.com]

There are lots of fast computers in space today. Spaceway has some pretty fast PPC CPUs. The Mars Exploration Rovers also have some fast ones. The real trick to getting a fast computer into space is in getting it radiation hardened and able to recover from upsets. The challenge comes from the fact that the die sizes keep getting smaller and smaller and this means that more gates would be affected by alpha particles. You're better off with a larger die size for space borne applications. Unfortunately, space is such a low demand market that the chip vendors don't produce anything suitable.

[antiRepublicrat]

You'd think a POWER4 would be perfect, with already thick gates and logic built into the system to automatically retry any job on one core that's not validated by the result in another core.

[Ladysmith]

Say what you want; for pure aesthetics, the Cray 2 was it hands down.

[Moonman62]

Their methods involve strategies such as making the computer fault-tolerant, or able to make an instant switch from a temporarily failing board to a functioning one. They also use algorithm-based techniques to detect and correct processing errors. “If one board is failing because of radiation, we can automatically go to another,” George said.

My little pet theory is that cognition evolved in biological brains for the purpose of fault tolerance.

[antiRepublicrat]

Cool, but it still doesn't beat one you can sit down on and eat your lunch.

[Moonman62]

HAL may soon be getting some company. But unlike the famous computer companion in Stanley Kubrick’s “2001: A Space Odyssey,” the first space-based supercomputer — so described because it will be by far the most powerful computer in space — is already nearing reality.

Hopefully, they won't ask the computer to lie.

[tundra1946]

Alpha particles are not the problem in space. See NASA website on Radiation and electronics for much information.

[Ladysmith]

Been there, done that. heh heh!

[SteveMcKing]

If networking principles were applied in space, there'd be a flood of great discoveries coming back to us.

For one example, imagine dispersing 1,000 little robots across mars... knowing very well that not one of them is expected to find much. The data returned by any single instrument would be unpredictable, creating an unattractive risk of acquiring random, useless information. However with enough probes taken as an amalgam, some very good studies could be done.

I think it may be the next big trend in space exploration.

Well, that and a long string of tragic Chinese moon-mission disasters.

[rintense]

Skynet is the virus!!!

[phantomworker]

My little pet theory is that cognition evolved in biological brains for the purpose of fault tolerance.

How about this one:

"There's no system foolproof enough to defeat a sufficiently great fool." --Edward Teller (father of the hydrogen bomb). ;)

[stripes1776]

My little pet theory is that cognition evolved in biological brains for the purpose of fault tolerance.

A pet theory would be correct. Purpose plays no role in the theory of evolution.

[annie laurie]

"There's no system foolproof enough to defeat a sufficiently great fool."

That's simultaneously profound ... and frightening ;-)

[Moonman62]

... for the advantage of fault tolerance. Does that make you happy?

[SFConservative]

There are two things going on.

One is a gradual degradation due to the absorbed total dose of radiation, which one can allow for by building and testing the devices properly and using the right amount of shielding so the hardware survives the planned mission duration.

The other concerns Single Event Upsets (SEU) and similar single event phenomena which temporarily cause an error. The approach to this is usually to have multiple computers perform the same task, and comparing the results in real time, and "majority voting" those results. The occasional oddball result from a computer experiencing an SEU is thus ignored.