[question] Distributing tasks accross a few comps.

Hey guys, I heard this can be done but I dont know how.

I have a mac mini with a 1.42 ghrz cpu and 512 ram, then under my

desk I have a pc with a 2.8 ghz pentium 4 cpu and 512 ram, and I also

have a pentium 3 500 mhz with 192 mb ram in a G3 case. Now I want

to netwosk all these comps somehow so that tasks are distrubuted

accross them. How should I go about doing this ? Is it even possible

without spending a huge amound of money ? The OS currently running

on both of the PCs is windows but I can install linux / unix if needed.

thx in advance,

Gabriel.

#293559

Highly doubt it (especially the "save money" part)

And why would you need a machine like that anyway?

#293561

If you're running mathematica, it can be done. Some apps support it, some dont. You'll have to be more specific...

#293562

Why would I need one ? What kind of a geek are you ????????? O_o

@ Phat-Lizard : Im just interested in it mainly for the speedyness of osx itself

on my mac and also photoshop.

#293563

You do realize that network-transportation is a couple of billions slower than 'normal' memory access. So the time that a `task` needs to be sent to another pc will be larger than if it's actually done on your cpu.

Parallelization can only be done efficient if tasks don't depend on each other. For example, computing all primes from 1 to 100000000000000000000000000000(a lot).

#293565

so in other words there is no way to use 3 comps to make one faster comp ?

#293568

Errm - if memory to cpu is "billions faster" than networked machines sharing tasks? what do you think, Hobbes?

It is probably possible; yes. Its stupid too because it wont be any faster (hell, not even the same speed)

#293577

too bad it isn't just three macs :P then you could have used Xgrid...

EDIT: Hmm it seems that u can get XGrid working on Linux and/or Unix.... but it seams like a lot of work.... And veeery hard :( Or to me it seems that way... http://unu.novajo.ca/simple/archives/000026.html

Oh... and XGrid is a simple system for setting up and using a cluster of OS X machines... Just so you know....

#293578

Can I have your old avatar? You know, the one from Ice Age, when you were iFido?

Thanks!

-NC

#293592

For example, computing all primes from 1 to 100000000000000000000000000000(a lot).

That's not that much... Mathematica will do that in a blink of an eye :)

The point is that networking is only usefull for things that are

  1. very cpu intensive
  2. take a long time

That's why the "find alien life" project by nasa (forgot the name) used several computers: each computer makes its own calculations, and just sends it to the server.

#293699

@Phat-Liz0rd: it was just an example ;) The idea is that a large task should be decomposed into larger sub-tasks that aren't dependant on each other. The find-alien-life-project is a perfect example for this. There's a large set of data (radiowaves?), and each pc can do calculations on a small subset. The subsets are not dependent on each other. If, for example, after every 10 calculations, the algorithm needed to wait on data from other subsets, things would slow down significally.

Of course, there are scientific grids and clusters, but those are highly optimized, and most of the time have special connections between the CPUs or share the same memorypool.

In short: Normal stuff is not gonna profit from this.

#293701

Well I dont want to be doing any large calculations or rendering or anything I just want the OS itself to be

faster. Photoshop aswell.

@ Nightcrawler : Yeah sure no probs :)

#293743

Well that wont happen :P

#293745

Buy more ram :D

#293869

=-o i got the mac i want up to 24,000 grand, that should have enough ram..

#294165

if you just wanna speed up the mini, get an external hard drive and boot from it, i boot mine from a 160GB firewire drive and it is soooo much faster

#297822