Rendering Distribution

[CosmicBear]

something like an internet-based renderfarm? hmmm, sounds good in theory.

but when i look at some of my files, i'm not sure, if it would be practically to upload them somewhere. with all the textures, etc. it easily goes up to a couple of hundreds of mbytes. and than there's the other way around. sending the rendered files back would do the same. uncompressed tif's can get quite huge (depending on the resolution of the image)
so imagine you have someone render a couple of images for you and then you'll have to send them back. would probably take some time (depending on your connection of course) and would produce alot of traffic on a webpage, wouldn't it? :shrug:

[handige_harrie]

Originally posted by CosmicBear
something like an internet-based renderfarm? hmmm, sounds good in theory.

but when i look at some of my files, i'm not sure, if it would be practically to upload them somewhere. with all the textures, etc. it easily goes up to a couple of hundreds of mbytes. and than there's the other way around. sending the rendered files back would do the same. uncompressed tif's can get quite huge (depending on the resolution of the image)
so imagine you have someone render a couple of images for you and then you'll have to send them back. would probably take some time (depending on your connection of course) and would produce alot of traffic on a webpage, wouldn't it? :shrug:

Size is indeed an issue if you are talking about textured scenes although I must say that 'a couple of hundreds of mbytes' is pretty extreme (IE: my logo scene is currently my largest C4D file at 4.4MB). But the bandwith/storage problem should be something we can solve.

Sending back the finished animation should be less of a problem if everyone compresses their part. The artist should define in advance what format should be used. Use of external programms like VirtualDub (for DivX) might be necessary.

If you use a password locked ftp, only participaters can download and upload, thus decreasing traffic a lot.
The public final animation will cause more traffic problems.

[flingster]

what issues are there around compatibility?
eg pc/mac path names etc, shaders installed, plugins possibly used?

then turnaround times?
eg how do you decide who renders which frames etc?

i like the idea...you could issue render points...for how much render time you used etc.

dunno its a toughie...we could do with someone coming up with something like the seti screen saver or dna one...which computed out of hours then uploaded results...eg web render...as opposed net render!!! now that would be funky.

i'm definitely feeling for ya though...
:shrug:

[handige_harrie]

Originally posted by flingster
[B]what issues are there around compatibility?
eg pc/mac path names etc, shaders installed, plugins possibly used?
That'll probably be the most difficult part, especially since you have people still on XL6, several versions of XL7, and R8.0 and 8.1. As for path names and shaders, including some extra stuff and some instructions should do the trick.

then turnaround times?
eg how do you decide who renders which frames etc?
By letting everyone run Cinebench, and distributing the frames according to their renderscore.

i like the idea...you could issue render points...for how much render time you used etc.
You mean 'earning' renderpoints by rendering other peoples files. Using these renderpoints to get your own stuff rendered by others? Might be a nice idea, but not really necessary.

dunno its a toughie...we could do with someone coming up with something like the seti screen saver or dna one...which computed out of hours then uploaded results...eg web render...as opposed net render!!! now that would be funky.
Wow :eek: didn't think about that. That's a few steps ahead, I don't see any of us doing that. Might be a very cool idea for Maxon to work out. And it could be based on Cinebench I think because it has already got a point system and a renderengine in a small application. The hard part will be the internet-distribution itself however.

I was more aiming for personal distribution 'case by case'. Just here on cg-talk as for as I'm concerned.

[squidinc]

26 hours isn't so bad really, couldn't you leave your pc on overnight and then the following day and a bit, but anyway doesn't net render work over the net? :) it only goes by IP addresses after all, just set up one machine as a server, someone with a static IP and decent connection ( on anything other than broadband it would be a total nightmare ) and then setup the IP in the clients, works sweet over a LAN, not sure about the net though

[LucentDreams]

there is an online C4D rendering service for anyone interested. Maybe Adam has the link on hand, I"ll try to find it at postforum.

[squidinc]

these guys?

http://www.renderking.com/

[LucentDreams]

thats the one

[Pate]

Originally posted by handige_harrie
Rendering one test-frame took about 3min30sec*. Rendering the whole animation of 450frames would take > 26 hours :D.

Radiosity or no radiosity: I am sure these kind of rendertimes will sound familiar with some of you ;).
[/B]

Oh, I wish... Currently my project renders WITHOUT RADIOSITY and WITHOUT SCENE MOTION BLUR on a P4 2.4GHz machine about an hour a frame, and this is after considerable optimizing and removing all non-necessary objects and features. My first test renders were about 3 hours a frame. I just finished the first 200-frame test clip after about a week of rendering. It has so many errors it is not presentable, but as a test render it was a success.

Still, I haven't yet lost hope about rendering the movie with radiosity and scene motion blur, I'm just not sure if I live long enough.. Let's see, if it takes about 30 hours a frame, how long an animation I can create in forty years, assuming I can upgrade my PC to two times faster every two years? :)

Anyways, what I meant to say, is that I consider 4 minutes a frame really fast rendering!

Some type of a web rendering system would be a neat idea, though!

Pate

[handige_harrie]

Originally posted by squidinc
26 hours isn't so bad really, couldn't you leave your pc on overnight and then the following day and a bit, but anyway doesn't net render work over the net? :) it only goes by IP addresses after all, just set up one machine as a server, someone with a static IP and decent connection ( on anything other than broadband it would be a total nightmare ) and then setup the IP in the clients, works sweet over a LAN, not sure about the net though

26hours is not really a problem, I know. And I managed to cut down rendertime (and get better radiosity results :D) to < 7 hours. I am currently rendering the last 300 frames.

NET render over the web is possible, at least the manual says so. The problems are:
-not everyone has NET render.
-everyone is bound to the server, they must all render parallel to eachother and with people all over the globe that's not really practical. I am not 100% sure about this, but it sounds logical to me. And I also don't know what happens if a connection is (temporarily) lost.
-My ISP gives dynamic IP adresses (external IP changes over time), so it is tricky to say the least.
-Everyone has to find out how NET render works, and I am sure we will encounter quite some problems the first times.

If you just distribute the file with a given number of frames to render, everyone is free to render it when they want (within a given period of time of course).

Point is, I don't think NET render will be a great succes.

Originally posted by squidinc
these guys?

http://www.renderking.com/

Well that's really nice if you have to do a lot of rendering for a paid job or a serious demoreel, the price isn't too bad.

But surely I am not going to pay money for rendering my own logo animation :p. And I think most of us cannot justify paying money for rendering their animation. Besides that, this is also an initiative to help eachother out.

Originally posted by Pate
Oh, I wish... Currently my project renders WITHOUT RADIOSITY and WITHOUT SCENE MOTION BLUR on a P4 2.4GHz machine about an hour a frame, and this is after considerable optimizing and removing all non-necessary objects and features. My first test renders were about 3 hours a frame. I just finished the first 200-frame test clip after about a week of rendering. It has so many errors it is not presentable, but as a test render it was a success.

Anyways, what I meant to say, is that I consider 4 minutes a frame really fast rendering!


Well my 3.5min/frame was just a small (and unconvincing apparently :p) example.
Yours is much better ;). This is a perfect example of an animation that might not even get finished without the help of others. As far as I'm concerned you get the scene done (testing, testing, testing), share the files, and dedicate some frames to volunteers here on CG talk. I am willing to render for you.

[Pate]

Originally posted by handige_harrie
Well my 3.5min/frame was just a small (and unconvincing apparently :p) example.
Yours is much better ;).


Heh, well as a (bad) example I suppose so! :grin:


This is a perfect example of an animation that might not even get finished without the help of others. As far as I'm concerned you get the scene done (testing, testing, testing), share the files, and dedicate some frames to volunteers here on CG talk. I am willing to render for you. [/B]

Thanks for the offer, I might bring the subject up here when I am at that point! It would indeed be nice to get some rendering help for such time-consuming hobby/non-commercial projects. My scenefile is about 30MB and I have about 40MB of textures, which means I would need to work out the details about how to share this data...

I also have some free CPU cycles to spare when I am not rendering so I could also help render some other peoples scenes occasionally.

Pate

[RickBarrett]

Seems to me NET render is the best solution - no sense reinventing the wheel.

Someone with solid, fast access, a static IP and plenty of bandwidth would set up the server. Everyone else just needs the NET client, which can be shared freely and doesn't need a serial number (and I think installed off your C4D CD). AFAIK, the client doesn't need a static IP.

Ideally the NET server would have FTP access so users can upload projects and download frames quicker. And an auto-archive function for finished frames would be handy so you could download 1 ZIP file rather than 1,000 TIFFs.

The only item left would be the management of who gets to render when and whatnot. It may be possible to build some web-based management around the server (seems like someone was working on that). I've done some similar stuff with a WAP-enabled version of NET. I'd be happy to help with this project where I can.

- Rick -

[dann_stubbs]

hi,

i dont' come in here often but with what postforum is turning into i might make my occasional stop here rather then there.

drop me an email and i'll set up an account for you so you can get your anim rendered.

the farm has plenty of spare cycles and your job should be done in a few hours.

i've had people upload over a gig of source files here using my FTP server and download about the same finished frames. someone just rendered a 6500 fram anim that took just under 24 hours. the small time to upload/download is far, far less then the time it would have taken to render.

to save time you can set your FTP client to auto log in every 30 minutes or couple hours or whatever and have it download your finished frames so when you get in you have your stuff on your desktop. really simple to do.

i like your thoughts on a distributed thing, i'm not sure it would work though. bandwidth issues along with constantly dropping and adding clients would be problematic as well as any uploads to to every CPU out there, that is why here i have the user upload to my server and then from there it is all 100base switched to the farm nodes.

the sad thing i feel reading this thread is when i started renderking last july i had it at $30 for a month of unlimited access to the farm and it attracted 3 people. i was pretty suprised at the low interest, i rasied the price to make it worth the electricity to leave them on and attracted a whole different user and it has been steady useage since - but can now justify leaving the farm up 24/7 and slowly adding improvements to it.

so drop me an email, i'll get it rendered for you easy... when you get some freelance and can justify doing a commercial render - just consider my farm as an option.

dann

[JIII]

I personally think it is really cool that dan stubbs is giving away render time.

That is really nice of you dude. two thumbs up, and a pat of the back for giving this forum an even better sense of community.

[Pate]

I'd like to thank dann for the offer as well!

I might even be able to pay something for access to RenderKing, that $30 a month does not sound that bad.. I had assumed you'd have to pay a lot of money for an access to a render farm!

When I get my scene finished (might take months still, I'm doing a VERY complex scene) I'll be sure to contact you dann!

Thanks!

Pate

[JIII]

hate to burst your bubble pate but his prices are about 11 times 30 bucks now.

so it is totally worth it to get some free time.

[handige_harrie]

Originally posted by dann_stubbs
hi,

i dont' come in here often but with what postforum is turning into i might make my occasional stop here rather then there.

drop me an email and i'll set up an account for you so you can get your anim rendered.

the farm has plenty of spare cycles and your job should be done in a few hours.

i've had people upload over a gig of source files here using my FTP server and download about the same finished frames. someone just rendered a 6500 fram anim that took just under 24 hours. the small time to upload/download is far, far less then the time it would have taken to render.

to save time you can set your FTP client to auto log in every 30 minutes or couple hours or whatever and have it download your finished frames so when you get in you have your stuff on your desktop. really simple to do.

i like your thoughts on a distributed thing, i'm not sure it would work though. bandwidth issues along with constantly dropping and adding clients would be problematic as well as any uploads to to every CPU out there, that is why here i have the user upload to my server and then from there it is all 100base switched to the farm nodes.

the sad thing i feel reading this thread is when i started renderking last july i had it at $30 for a month of unlimited access to the farm and it attracted 3 people. i was pretty suprised at the low interest, i rasied the price to make it worth the electricity to leave them on and attracted a whole different user and it has been steady useage since - but can now justify leaving the farm up 24/7 and slowly adding improvements to it.

so drop me an email, i'll get it rendered for you easy... when you get some freelance and can justify doing a commercial render - just consider my farm as an option.

dann

Well that's a real nice offer :thumbsup::applause: but I finished rendering my short animation yesterday (total of 18 rendered seconds in 9 hours :p). So I won't be needing the rendertime. If you still want to help someone out, please let Pate make use of your spare cycles.

I know letting you do the rendering is more efficient, my idea was meant as a thought for 'hobbyists' to help eachother out.

What do you mean by 'the sad thing i feel reading this thread'? The reminder of almost no-one paying 30$? Anyway I am glad that now you do have customers.


NET render over the web is not the easiest solution imho, although it would be very nice.

I still believe in my own idea of distributing the files and letting people render when they have the spare time. It has been done before with a group project and it worked like a charm. There is no such thing as constantly dropping and adding clients.

Bandwith shouldn't be an issue with most scenes I think, and if it is, there must be a simple peer-to-peer solution.

I know I am stubborn :wip:

[Pate]

Originally posted by JIII
hate to burst your bubble pate but his prices are about 11 times 30 bucks now.

Ouch.. Oh well, when something sounds too good to be true, it usually is. :D

Pate

[flingster]

dann: very nice offer...excellent sense of community spirit. its wierd that the price change effectively changed your business. for any small scale artist you would think the offer of 30 bucks a month would be irresistable...:shrug: . why not also think about for non commercial work...offering the same service as 30bucks...then again it might affect your commercial side..which obviously would be a bad thing. just a thought. sounds like an excellent service you got there...wasn't even aware it existed...
:thumbsup:

[dann_stubbs]

Originally posted by handige_harrie
Well that's a real nice offer :thumbsup::applause: but I finished rendering my short animation yesterday (total of 18 rendered seconds in 9 hours :p). So I won't be needing the rendertime. If you still want to help someone out, please let Pate make use of your spare cycles.

I know letting you do the rendering is more efficient, my idea was meant as a thought for 'hobbyists' to help eachother out.

What do you mean by 'the sad thing i feel reading this thread'? The reminder of almost no-one paying 30$? Anyway I am glad that now you do have customers.


NET render over the web is not the easiest solution imho, although it would be very nice.

I still believe in my own idea of distributing the files and letting people render when they have the spare time. It has been done before with a group project and it worked like a charm. There is no such thing as constantly dropping and adding clients.

Bandwith shouldn't be an issue with most scenes I think, and if it is, there must be a simple peer-to-peer solution.

I know I am stubborn :wip:

if you go to my NEWS page you can see all the way back to when i first announced renderking as a way to share my own personal farm. i was focused on a benefit for hobby/single users - yet the idea just died on the vine. it appears in hindsight that there is less need for a cheap render solution out there - maybe since a true "hobbyist" isn't encumbered by deadlines they can really afford to wait for the render... i'm not totally sure the reasons it didnt' work.

it costs about $200 a month electricity to leave these all on 24/7 as well as there are additional wear and tear issues (like when a powersupply burns out at $70 to replace or a motherboard burns out like happend two days ago big smoldering white blob where some chip was) these all cost money to replace.

i was trying to offer my farm to everybody with such a low price and some small income to help pay the power bill and upkeep in return. but funny it started all these complaints from hobby type users about how long it would take to get their job rendered, etc. and even more funny these people are the same to never use the farm - i even offered many a free month to test it out - not one ever accepted - they just wanted to rip on something. (sort of like the people who constantly argue about software but obviously never really use any of them) the sad part is that renderking was born in exactly the same "thought" as your idea of distributed rendering.

the problems with the distributed method are the server will get killed with trying to serve textures to a few dozen CPU's scattered across the world. each mb will have to be sent to each machine. that bandwidth will add up FAST - as well as the timeout on the server will have to be high enough to let large textures transfer on possibly slow connections before giving an error. my comment about adding and dropping clients is valid. these users if they use their computers will be quiting and starting (not including the occasional crash) their net render client from time to time if they want to actually work on their computer which will add or drop on the server. if a job is rendering that means a dropped client and the server will now have to redistribute.

bandwith may not be an issue as long as your scene and textures are under a couple megabytes, but users here as i said have projects that average in the hundreds up to gigs of data - that would never work over a distributed setup for reasons i said above, but i guess if projects are that big they may not be looking for free usage to render so maybe a moot point.

i am thinking of a way to offer an occassional "free" render for non-commerical work under a certain frame limit. of course the paying users will still have the higher priority in the render queue. but still may offer a "hobbyist" some renderpower options. all i can say is send me an email if/when you have a need - give me specifics on the render and i'll see how it goes in getting them added to the queue here.

dann

[handige_harrie]

Originally posted by dann_stubbs
i was trying to offer my farm to everybody with such a low price and some small income to help pay the power bill and upkeep in return. but funny it started all these complaints from hobby type users about how long it would take to get their job rendered, etc. and even more funny these people are the same to never use the farm - i even offered many a free month to test it out - not one ever accepted - they just wanted to rip on something. (sort of like the people who constantly argue about software but obviously never really use any of them) the sad part is that renderking was born in exactly the same "thought" as your idea of distributed rendering.
Ok, now I understand. It is a shame indeed. I don't get why someone would decline your offer of free farm usage for a month :eek:.

the problems with the distributed method are the server will get killed with trying to serve textures to a few dozen CPU's scattered across the world. each mb will have to be sent to each machine. that bandwidth will add up FAST - as well as the timeout on the server will have to be high enough to let large textures transfer on possibly slow connections before giving an error. my comment about adding and dropping clients is valid. these users if they use their computers will be quiting and starting (not including the occasional crash) their net render client from time to time if they want to actually work on their computer which will add or drop on the server. if a job is rendering that means a dropped client and the server will now have to redistribute.

That's exactly why I think NET render will not work!

Let me give a example of a group project that did distribute the rendering the way I propose:


The 'artist' finishes a project, ready to be rendered.
The artist makes a topic here, to get people who will help him render the animation (lets say 10 people volunteer).
The artists gets information about everyones pc/mac (preferably with Cinebench results) and assigns frames accordingly to everyone.
The artist decides how long the renderers have to complete their part. Lets say two weeks.
The artist decides what the final output has to be (compressed, but still good quality).
The artist puts his zipped C4D-file and textures on a password locked server.
The artist sends the 10 renderers the password via email.
The renderers download the files.
Everyone renders their files within the two weeks.
The renderers upload their compressed part of the animation to the server (again password locked).
The artist downloads the separate parts and puts them together (using VirtualDub for instance).


In this case there is no such thing as constantly dropping and adding clients ;).


bandwith may not be an issue as long as your scene and textures are under a couple megabytes, but users here as i said have projects that average in the hundreds up to gigs of data - that would never work over a distributed setup for reasons i said above, but i guess if projects are that big they may not be looking for free usage to render so maybe a moot point.


Yes, I also think if someone made a scene that large, they should better contact you ;).

Lets take the 10 people example. Lets say the file is 100MB and the final animation is 500MB.

10x100MB = 1GB traffic
10x50MB = 0.5GB traffic

I personally think this is quite a large project then, and the traffic isn't too bad. I don't know if tripod/geocities permits password usage. If so, these might even be an option!
And if server bandwith is a problem maybe peer-to-peer is an option.

The final result posted here on cg-talk will probably cause more traffic.


i am thinking of a way to offer an occassional "free" render for non-commerical work under a certain frame limit. of course the paying users will still have the higher priority in the render queue. but still may offer a "hobbyist" some renderpower options. all i can say is send me an email if/when you have a need - give me specifics on the render and i'll see how it goes in getting them added to the queue here.

dann

I will keep that in mind, it's a very generous offer :).

[dann_stubbs]

Originally posted by handige_harrie
Let me give a example of a group project that did distribute the rendering the way I propose:


The 'artist' finishes a project, ready to be rendered.
The artist makes a topic here, to get people who will help him render the animation (lets say 10 people volunteer).
The artists gets information about everyones pc/mac (preferably with Cinebench results) and assigns frames accordingly to everyone.
The artist decides how long the renderers have to complete their part. Lets say two weeks.
The artist decides what the final output has to be (compressed, but still good quality).
The artist puts his zipped C4D-file and textures on a password locked server.
The artist sends the 10 renderers the password via email.
The renderers download the files.
Everyone renders their files within the two weeks.
The renderers upload their compressed part of the animation to the server (again password locked).
The artist downloads the separate parts and puts them together (using VirtualDub for instance).


In this case there is no such thing as constantly dropping and adding clients ;).



ahh, yes that does circumvent that issue, but in two weeks most people could get a big render done themselves in many cases. the only other issues are the time to control and arrange all these transactions and arrangements (something i don't have) but i guess in other circumstances mabye not a problem. as well as "true" security of your art/files.

the biggest concern i would have is common processor architecture. especially in raydiosity or math oriented processes. each processor type has different math and rounding characteristics. (P4, P3, AMD, mac) so each chip could create variations in the sequence of renders. again maybe not always a problem with all renders but definately an issue to be watched.

so you spend two weeks arranging this large render and upon final completion you have variations and differences in the final frames due to processor differences. maybe to have to start the whole process over again - but this time adding another limit on processor type for people who can contribute. it can get quite messy on the worst case side of it.

again maybe not a problem for all but it would be a problem for many - i think that is where the free farm has some hurdles to work out. just something to be aware and maybe concerned about. but i wish you luck with the efforts to get a 'public free farm' up. i'll look for info regarding the efforts and i'll help out if and when i can...

dann

[handige_harrie]

Originally posted by dann_stubbs
ahh, yes that does circumvent that issue, but in two weeks most people could get a big render done themselves in many cases. the only other issues are the time to control and arrange all these transactions and arrangements (something i don't have) but i guess in other circumstances mabye not a problem. as well as "true" security of your art/files.
Two weeks is quite long I know, one or two days would de ideal, but it would be a miracle if everyone had the time available on the same day. Besides that there are people who rather not leave their pc on during the night for various reasons (I'm one of them). If I had to render something for two weeks straight I would not be able to use my pc (effectively) for other things. I'd go mad :D.
In two weeks everyone can spread the renderjob the way they want.

Regardless the way you look at it, it still remains a fact that 10 can render more than 1 in the same period of time.

Managing this will require some time, but it is the choice the artist makes. Render the whole thing himself, or take the time to distribute it. In some cases (like Pate's) doing it all yourself is just not feasible. And that's the reason for attempting this in the first place.

As for security, first of all you have to trust the people that help you out in rendering. Only these people should have acces to the files (password locked server). Still this is no guarantee the files wont leak out through the render volunteers but if it leaks out at least you know who might have been the leak.
If the artist is not comfortable with this he shouldn't distribute the rendering. He'd better contact you or render it himself. It's his choice.

the biggest concern i would have is common processor architecture. especially in raydiosity or math oriented processes. each processor type has different math and rounding characteristics. (P4, P3, AMD, mac) so each chip could create variations in the sequence of renders. again maybe not always a problem with all renders but definately an issue to be watched.
I know what you are talking about. However to me it would be a big surprise is these rounding flaws are visible in the final result. But if we try, we don't know. I will make a small scene using radiosity, caustics, raytrace and dynamics. If this doesn't create visible differences I am pretty sure nothing will.

again maybe not a problem for all but it would be a problem for many - i think that is where the free farm has some hurdles to work out. just something to be aware and maybe concerned about. but i wish you luck with the efforts to get a 'public free farm' up. i'll look for info regarding the efforts and i'll help out if and when i can...
dann
There is always a risk of the result not being satisfactory, but that has nothing to do with distributing the rendering (that is: if the cpu-differences don't cause any trouble).

Your help (and that of others just as well) is always welcome :).

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.