Distributed Movie Rendering

[TEK NINJA]

Some time ago I created a thread here about my distributed image renderer. It was not very efficient, but it worked as planned. This time, a friend and I have dediced to concretize what seemed like "the next logical step". We have decided to create a distributed 3D movie renderer, which will operate over the internet. This renderer will use photon mapping to render frames and will be much more efficient than my experimental distributed renderer. Our goal is to be able to render ~15 minute movies in 640x480 in 48 hours or less (thats 21600 frames at 24 FPS).

To do this, we will run a central rendering server, which will assign work units, in the form of individual frames, to the clients. The clients will be able to run the client either in the tray, or as a screensaver, or both. The cool thing is that the clients will be able to see what is being rendered... We will also have statistics on our website detailing the performance of the contributors, how they rank overall, etc. People will be able to sign in multiple machines over their account as well.

Whats the point? The point is to allow amateurs to create their own animated CGI movies. Eventually, we would like to render some kind of small sci-fi series and such, and allow people to download the movies off of our website once they are rendered. The system will provide means for creative expression of that form. We will encourage artists to create their own animations so we can render them.

My question is: Who would be interested? Obviously its a bit early. It will take us at least 3 months to have some distributed rendering going on, but I wanted to know how much interest there is. When we get the system working, we might actually post about ourselves on slashdot to get as much help as we can.

I am, of course, open to questions and feedback about this.

 

[Ch1m3r4]

I guess the question is how many computers do you have folding for the [H]orde right now?

I mean, we're always more than happy to help out people, but even more so if that guy has 10Ghz of folding power going for him

Especially now that those Aussies are slowing down...

 

[sharp]

medical DC is it for me (long term) but if you need a tester for you client or whatever I'll run it for you.

sharp

 

[ByteDown[H]ard]

Sounds interesting . I'd be willing to donate a little time to the cause. Any idea what the min machine specs would be for something like this??

 

[TEK NINJA]

The minimum specs would be in the range of:

- 500 MHz processor or faster
- 256 MB RAM or more
- 300 MB of hard drive space
- Broadband internet connection

Nothing extra fancy. We won't use 3D graphics APIs either, so no need for an ultra fancy videocard, a good old 2 MB PCI videocard (or none at all) will do. Of course, the faster your CPU, the faster this will run, and we will support multi-processor systems as well.

 

[SoL]

I knew you would come up with another project, good to see. I was quite impressed with your last effort. For those who missed it, have a look at this thread.

Keep us updated, I'll donate a machine or two.

 

[Bodega]

does it use a significant amout more memory than f@h? 128 is all i have in alot of my folding boxen

 

[TEK NINJA]

I knew you would come up with another project, good to see. I was quite impressed with your last effort. For those who missed it, have a look at this thread.

Keep us updated, I'll donate a machine or two.

I had to make something better But this time I am now alone, I have a good friend with at least as much skills as I have. We will hopefully have some actual distributed rendering going on in a few months.

does it use a significant amout more memory than f@h? 128 is all i have in alot of my folding boxen

The system has to load all of the scene data in memory to render, so it really depends on the complexity of the scene. I suppose at the start, with simple scenes, it won't be a problem, and probably won't take more than 30-40 MBs of RAM... If we are to have very complex scenes later on though, it might take alot more than that.

 

[Bodega]

The system has to load all of the scene data in memory to render, so it really depends on the complexity of the scene. I suppose at the start, with simple scenes, it won't be a problem, and probably won't take more than 30-40 MBs of RAM... If we are to have very complex scenes later on though, it might take alot more than that.

I think thats one thing that will be a problem with getting alot of people to run your DC project. In this forum most of us run f@h, which takes up very very little ram. This allows whoever is running it to still use the computer as normal.

 

[Carnival Forces]

yeah, i'd contribute some time, but i think it's folding for me on a long-term scale...although self-expression is damn important too...

 

[Chugiak]

I have no personal desire to run the project, but am happy to offer criticisms!

Seriously though, I hope you can get it off the ground. But since anyone here who would be contributing would do so knowing that it would be that many fewer cycles for medical research, I doubt you'll find a lot of folks in this forum jumping on board. Nonetheless I'm also sure lots of geeks here can offer some insight into what makes a successful distributed computing project.

I'll add on to the comments about RAM above: assuming you want wide distribution, you're going to have lots of different degrees of performance. You'll need to either make work units deliberately small so they can run on most everything or else devise a system of allocating work units based on the capabilities of the end system.

Good luck!

 

[unhappy_mage]

I have no personal desire to run the project, but am happy to offer criticisms!

me too

Why not use POVray? I know it removes the cool factor of writing your own raytracer (congrats, by the way; that's no mean feat), but it's standardized, people who write scenes don't have to learn a new modeling language, it's optimized, etc. Cool project nonetheless.

 

[TEK NINJA]

First of all sorry for not answering for a while, my CS exams keep me busy.

me too

Why not use POVray? I know it removes the cool factor of writing your own raytracer (congrats, by the way; that's no mean feat), but it's standardized, people who write scenes don't have to learn a new modeling language, it's optimized, etc. Cool project nonetheless.

Well there are a few reasons. First of all, from my experience, its very often easier to create your own product than trying to adapt someone else's. It could be quite hard to add the features we want, especially since they use some hacked C code. Also, POVray doesn't do that good of a rendering (its decent, but not movie quality). Finally, it won't matter to the artists. We are going to create our own 3DS max exporters.

There's also the non-technical reason that this project is also meant to teach us something. I don't like getting into projects where I won't learn a thing. Thats partly why I have started my own initiative to create a 3D game engine (www.xgameproject.com) instead of creating a Half-Life modification, for example.

As for the RAM usage. There's not much we can do. We can't store data on the hard drive, as that would be much too slow. Raytracers are also not very predictable when it comes to memory usage, so we can't really say "load the textures only for the nearby objects"... That just won't work.We will probably not aim at F@H people either way. I know you guys don't want to let down folding, and I have no intention of encouraging you to do so. I was rather thinking of posting about our project on slashdot.com, gamedev.net, flipcode.com and various other programming sites. I will try to appeal to people who have recent workstations and will be interested in the "cool factor". The problem, however, is that those people might be FBNFs. We will have to do something to keep them addicted. I was thinking of keeping lotso stats about the top clients, and eventually setup competitions to win stuff, that way, they will do anything to increase their rank in our TOP100 list