Render Farm Poll

http://www.bullyrender.com/img/br_screenshot_small.jpg
–edit–
Here’s a screenshot from a test spiffyandy is helping me with. In the preview screen you can see thumbnails of frames as they render. Each frame of this animation takes 5-6 minutes to render.
–edit–

I have recently been working on blender render farm software for a project of mine. The goals are for it to be highly scalable (possibly thousands of nodes), easy to use, inexpensive, and put no hardware requirement on the user (i.e. you don’t have to own 40 computers). More details are at the bottom of the post.

Now I’m at a crossroads. I can complete the project for my own personal use - end of story. Or I can make it a multi-user environment capable of handling many full-size render jobs at once. That would entail a good bit more work, but I’m willing to do it. The catch is, I’m not going to spend countless hours making it multi-user if no one would really use it. So the question I pose to you is:

Do you need a render farm? Would you use my render farm? If so, how often would you use it on what size projects?

Please be honest, the answers will determine if I open this project up for community use, or keep it private.

If you’re still reading, you’re probably interested in the details. Here they are:

Workflow:
You log in to a web interface where you upload a .blend file and wait for a few preview images to be rendered. An estimate is then calculated as to the total time and cost involved to render the entire scene. If you approve the job, it will start rendering immediately and hopefully be done within an hour or two. You could start downloading frames as soon as they were available. All you would need is a web browser and a .blend file.

Infrastructure:
The software is built on the popular Amazon EC2 and S3 infrastructure where you can rent computer processing power. When you start a render job, the farm requests more computers (render nodes) be allocated. When they’re done with your animation, they’re released. The configuration would initially be aimed at animations using the blender internal render. Conceivably, it could handle yafray or single image (mosaic) renders if there was demand for it, but that’s not on the radar now.

Cost:
The render farm would be cheap, but not free. Sorry. I rent the processing time, so if I rendered your animations for free, it would cost me money! I think the final price would be somewhere around 10 - 15 cents per GHz-Hour, but that may be adjusted for actual usage patterns. As far as I can tell, that’s a good rate compared to existing render farms (and thus the impetus for creating this software to begin with - to save me money on my animations). There are also file transfer fees and other miscellaneous fees, but the 10-15 cent number has been slightly inflated to compensate for that already. Payment would be made after a job was estimated, but before it actually ran.

Status:
I currently have a small proof of concept running that could be ramped up to a secure, multi-user environment. But it will take a while, in the meantime, I would need a few guinea pigs with medium sized animations to render that weren’t on a time crunch (things can and do go wrong during development, so you’d have to be able to roll with the punches)

Personally, I’d be more interested in the software itself than in purchasing time on someone’s render farm who’s renting the Ghz to begin with. A lot of people have (legal) after-hours access to their office’s computers and would love to be able to set something like this up for Blender.

I understand that. I initially started this project as an easy to install local render farm. I eventually changed my overall design because I own 5 computers, 2 of which are good. Even with excellent renderfarm software, my computers can’t handle big render jobs in a timely manner. That’s when I looked at renting time on a big render farm, like respower or render planet. They have hundreds or thousands of computers, but they’re a little pricey. The design goal here was a render farm without hardware requirements. The software relies explicitly on the Amazon EC2 architecture, and therefore can’t be installed on just any computer. This allows an interface like the bigger farms, and times on par with the bigger farms (maybe even faster) but at a lower cost.

My easy-to-install local render farm is 75% done and abandoned - at least for now. Perhaps you’d be interested in having that code, although there are several options if you are looking for local render farms. This implementation ran from a php web server which is very easy to install these days. That’s all you need is to install the server and drop in the code. Then you can upload jobs and run them. Render nodes only have to run the render script. Since all the communication is done over http, the nodes don’t even have to be on the same network and firewalls aren’t a problem. It was designed exactly for the type of situation you described, where you have access to several heterogeneous computers and would like to take advantage of their idle time without a big messy install.

My position exactly, but I’d still give War’s setup a go until that option becomes available.

I would love that, I know that it takes a long arsed time for them to render on my computer, I would be willing to supply an animation, I am just starting on them, but I would be happy to give you one to test it with.

Well, the results haven’t been staggering in sheer numbers of people replying, but the votes leaned toward the useful category so I’m giving this a shot. I bought a domain name. The render farm UI is now located at www.bullyrender.com and has an initial copy of my software.

I need testers. I have animations to run through the farm, but having a variety is a better test. Test animations should use blender internal renderer, not be too terribly intense (as I will be footing the bill for the first couple through) and not be on a strict timetable. Most of this seems to work, but there could be delays if I have to change for fix anything. Any of you who’ve written software know how often you create a program that never needs fixing or changing, so please, for the initial testing phase, don’t submit jobs you need done in an hour, it probably won’t happen.

If you would like to help with a test, please PM me or contact me through the website and I’ll get you all set up. It’s designed to be quite easy.

As of now, I let long renders run overnight, or over a weekend on computers that would otherwise be idle. Eventually, I’d be interested in a render farm.

I don’t understand the cost. I would think in terms of, say, a two minute animation where each frame took maybe 20 minutes to render. So thats 40+ days of rendering. I can split it up by hand to render on several computers, so the total calendar time might be a week or two.

So, given this scenario, if your renderfarm could get me the finished product back in a day or two, I’d gladly pay five or ten dollars for the service. On the other hand, if the service cost fifty to a hundred dollars, I’d spend the two weeks instead.

What I don’t get is how to translate the 10 to 15 cents per GigaHz hour into what it would cost for one of my “typical” projects.

Perhaps a rough cost estimator would be a useful addition to the site - keeping in mind there are many factors other than raw processor clock speed that affect total render time.

To make the math easy, lets say you have a 1.0 GHz processor. A frame takes 20 minutes (or 0.33 hours) to render on your computer.

0.33 hours * 1.0 GHz = 0.33 GHz hours per frame
at 15 cents per GHz, that’s about 5 cents per frame

If you render 30 frames per second for a total of 2 minutes of animation…
30 fps * 60 seconds/minute = 1800 frames per minute
2 minutes * 1800 frames/minute = 3600 frames

3600 * $0.05 = $180 for a total of 1200 GHz hours, or 50 days worth of your 1.0 GHz computer running non-stop.

Interestingly, if you do run this on your home computer, and you use between 1/3 and 1/2 of a kilowatt when rendering, which is typical, and you pay $0.10 per kilowatt-hour, which is also typical, you will spend between $40 and $60 on electricity alone.

Doing all this math makes me glad I’m not a Maya/MentalRay/3DS…etc user. When browsing renderfarm prices I noticed prices from $0.40 to $1.00 per GHz hour. At those prices, your render job would cost from $480 to $1200!

So to answer your question, that’s $180 - $50 (your own electricity) = $120 more you’d be paying to have the animation rendered for you. There is definitely a cost involved and it may be more than you’re willing to pay. I’m willing to pay because I make money on my work and this is the cheapest way I can get something done immediately. The advantage to the render farm is of course speed and convenience. You could have your animation done in an hour or two without worrying about segfaults and when you can use your computer instead of waiting weeks or more.

Hope this helps

Thanks for the info, War. The part of LA I live in has socialized electricity, so the price isn’t quiet that high, but it’s a good analysis even so. As a hobbyist, I’d probably spend the extra couple of hundred bucks replacing disk drives or power supplies on broken computers, than spending it on a rendering service. If I had clients, the picture would be entirely different, of course, since in that world, time is money.

I’d expect as more users moved into commercial Blender use, the demand for the service would increase. So it’s probably a very worthwhile project for you, although it may not be a big revenue stream right away.

hey war, was I supposed to upload the file, or are you going to do so? If I need to, I shall be able to upload it come monday, sorry for the delay

spiffyandy,

I’d like if you could upload the file so you can see if the user interface is easy enough to use. If you don’t have time, I’d be happy to do it though. I’m considering writing a python plugin so you can upload your current file directly to the renderfarm and download the images to automatically from within Blender at the touch of a button.
That will come later though, I do have a full time job after all.

i’d like to try the service… if it’s really cheap :smiley:

so how do you do this certain file type for the render farm?
Also you should have the time settings be adjustable for people in different time zones. That way people can more actively see how long a render.