SETI@home & RC5

abh writes "The SETI@home project is now sending new (fresh) work units, after having spent a few weeks in a rut sending the same 2 days worth of observations repeatedly. Read the announcement " As well, we were sent word from a reader that we've lost the #1 position at RC5. Head over and sign up!

Comments

Re:Use of other processors besides main CPU

[Leapfrog]

The video hardware in your computer is highly optimized for doing video things. Stuff like matrix transforms and rasterisation can be made very fast in silicon, but bog down a lot in software. So, SGI has implemented chips for this specific purpose. Typically, the way these very fast bits are accessed is through the OpenGL library, which provides a convenient way for 3d programmers to, well, program 3d apps.

Unfortunately, because the hardware is so highly optimized for making pretty pictures, it is pretty much useless for doing other things, like cracking encryption.

Theoretically, it is possible. Most 3d hardware needs to have things like add, multiply, etc in order to do the other stuff. But depending on the implementation of the hardware, it could be very difficult to access these parts, and even the access itself would probably be enough to slow things down quite a bit. It's akin to asking your sound card to act like a modem. While this is possible (see soundmodem drivers for linux kernel) it's not necessarily what the hardware is good at. (soundmodem driver is limited to 9600 baud max, using X.25 packet protocol, half duplex.)

It would require a massive code effort to warp the rc5 client around to use the OpenGL library for things the OpenGL library was never meant to do, and it probably wouldn't be much of a win anyway due to the odd manipulations you would have to go through to make a task-specific processor do things that its not specific to doing.

The SETI@Home announcement link is wrong

[Pulsar]

The link should be http://setiathome.ssl.berkeley.edu/tech_news.html which takes you to the announcements. The link Hemos posted is to the main page of the SETI@Home site which just carries a 1 line explanation.

If you read the actual announcements, you'll see it was only one day when the same 2 days worth of work units (actually not even two complete days, just 115 work units from 2 different days) were being sent, then the project moved to about 30k units, which were all that were avaible at the time. Now several months worth of data is avaible. They go into detail as to problems they have had and their solutions on this page, it's rather interesting...

Re:Here's a question

[pfournier]

That's why Cosm will use signatures... you either sign yourself every code you trust (and you'll have the source available for that checkup) or you allow code signed by other persons you trust. You can then be sure your clients will only run projects you want. Come take a look at Cosm, stop by the channel #Cosm on EFnet or help us code it.. :)

---

Here's a question

[for(;;);]

Could this stuff be put to a *really* practical use? Surely some math professors out there have some linear algebra number crunching to do, or some physicist has weather modelling simulations to make, or the like. Not that generating primes and scanning gads of radio waves for intelligent signs isn't nice and all, but it seems to me that gargantuan amounts of parallelized computation time could really help out researchers on tight budgets.

Is there an easy process for researchers to utilize this stuff? If not, it might be a Good Thing to set up.

(It'd have to be secure, I suppose, and in a best-case scenario could involve timesharing without administrative hassle...)

Re:Here's a question

[CComp]

This is why distributed.net is doing the RC5 thing. You think that whole effort is funded on the $10000 prize and that's the sole reason they're doing it? Nope, the point is to debug and test a distributed-work client that will let scientists and researchers do exactly what you want them to be able to do. And if it works like it's supposed to, the dist.net folks are gonna be able to retire early.

Too many people look at a thing and say, 'This should be used for X instead of Y, and right now! Change it! Change it!! Nownownow!!' Mebbe the intermediate steps are NECESSARY to get to where they want it, but no one cares about that. Gotta be instantaneous or nothing. Call it the McDonald's Syndrome.

It isn't going to happen overnight, but at least by supporting the distributed.net RC5 project, it will happen soon.

Well

[Shafik]

Yeah but Slashdot will topple Team EvangeList from overall 1st place at this rate in about a year (if the contest lasts that long). AnandTech, even if they double their rate it would take them 3 years to catch up.