BOINC Project idea : fractal@home

[ker2x]

I am currently exploring the possibility to create a BOINC project to render fractal on a very large scale (and exploring fractal that could take many months or year of cpu-time to render, or animation).
The documentation is here : http://boinc.berkeley.edu/trac/wiki/ProjectMain

My main goal is to work on buddhabrot rendering. But (if it works) we could do something more "universal" to render many kind of fractal with many kind of parameters.

Of course, to reach a greater popularity (and get much more client) the project should create awesome 3D (or 4,5, many-dimension) Fractal like the very popular mandelbulb.
Boinc also recognize & handle GPU computing too.

I think i can take care of the server-side stuff (100% sure for the sysadmin (being a sysadmin myself), not sure yet for the software part of the server).
and probably the client side stuff for buddhabroth rendering. For something more "universal" (and/or 3D) i probably can't.

One of my idea to create a "universal" fractal boinc client could be to simply use LUA.

This was my little brainstoring. It's your turn now

[cbuchner1]

Nice idea, but aren't buddhabrots are difficult to render in slices or pieces? So it's hard to split up the work.

I imagine you want to render a huge image in the many megapixel range (or even a Gigapixel image). So every user would have to keep all that in memory because most escape trajectories will somehow cover parts of the entire image.

Of course every user could throw away 90% of the computed trajectory coordinates and fill only a 10% slice of the image with any pixels that happen to fall into that slice (as an example). But that's wasting a lot of CPU cycles.

[ker2x]

Nice idea, but aren't buddhabrots are difficult to render in slices or pieces? So it's hard to split up the work.

I imagine you want to render a huge image in the many megapixel range (or even a Gigapixel image). So every user would have to keep all that in memory because most escape trajectories will somehow cover parts of the entire image.

Of course every user could throw away 90% of the computed trajectory coordinates and fill only a 10% slice of the image with any pixels that happen to fall into that slice (as an example). But that's wasting a lot of CPU cycles.

I use Monte-carlo method to compute mandelbrot.

Do a few millions of time :
1) choose random complex point
2) iterate and draw orbit
3) goto 1

a) each client could do that "1 millions time" processing.
b) the server retrieve thousands of "1 million orbit" map, add all maps together, normalize, render a single image.
c)
d) profit !

[KRAFTWERK]

This sounds cool  count me in if it gets real, I am on SETI@home already 

[Nahee_Enterprises]

I am currently exploring the possibility to create a BOINC project to render
fractal on a very large scale (and exploring fractal that could take many
months or year of cpu-time to render, or animation).

Sounds interesting, so please keep us informed as to your progress with this.   
 

[ker2x]

i'm reading the documentation. it won't be easy 

[Catelee2u]

Awesome idea! How is it progressing?

[pleasuretek]

You should look at how 'electric sheep' does it. Electric sheep renders flame fractals on distributed network while screensaver is active. Really amazing piece of software, already does some of what you are thinking about, nust not on BOINC.

[erstwhile]

Did anything ever come of this project? 
There's now an interesting bitcoin clone for BOINC called gridcoins that lets you split your cpu between BOINC and mining for gridcoins, so you can get paid (in gridcoins) for the CPU time you contribute.
A similar system could be really interesting for fractal rendering - you could lease your idle GPU's to people on BOINC or a similar network and get paid for it .