PDA

View Full Version : A dream: an OUR very big renderfarm with Lightnet:Cloud



Wilfrick
12-27-2011, 11:42 AM
Hi,

I hope to can explain this idea with my bad english...

There is a very good sw, Lightnet: Cloud, some months ago I used it and it works very well (http://www.joejustice.org/lightnet-cloud).
The programmer, Joe Justice, is a very good man and when I had some problems he helped me... if you think that the software is free you can understand that all his time spent to help me was a very big present.

I think.... why we don't build an our renderfarm? Need only a FTP and all us can join with our computers. Maybe NT can have the FTP and give this service to all LW users. We need only turn on lightnet when we go away from comp and it joins to render list.
Have you ready a scene for render? ok, send it to ftp (with lightnet) and all our free comps will render it.... more we are more the render will be fast.... you can save weeks! In one night you can have your frames rendered, frames then usually need 15 days to be ready, for example...
Of course need some rules (and programmers must put them inside the software), like a priority.... maybe who joins his comp for more than 5 hours per day is in CLASS A and his scenes are with high priority in render queue, who joins only 10 minutes per week is in CALSS Z and has a very low priority in render queue...
The scenes that require, for example, 12 GB ram don't must run on comps with only 4 GB...
The scenes that require some plugs must run only on comps that have these plugs (can't ask to people to buy plugins, but can ask to install free plugs, like DP, to join the "farm"...).
The scenes that require LW11 must run only on comp that have Lw11...
None can download from FTP (to stole) the scenes and the images of other people...
and so on....

I think that if NT builds this "toy" his users can have a really wonderful and incredible service... a superbig renderfarm for all Lw users.......

A dream....

I don't know if Lightnet:Cloud now give this possibility or need some very big features, but Joe wrote a great software and I think that he can do all to make this real.

Ok, I hope you can understand my english and sorry for errors.
Bye

ShadowMystic
12-27-2011, 02:29 PM
Interesting idea. Kind of like [email protected] Distributed computing power at work.

monovich
12-27-2011, 03:42 PM
who pays the power bills of all the people running constant renders on my computers? Me? Hmm...

Dexter2999
12-27-2011, 03:57 PM
who pays the power bills of all the people running constant renders on my computers? Me? Hmm...

Yeah, and when you are doing your renders on my computer who is paying for them?

It's not like the system is asking you to volunteer your computing power and never take advantage of the network. And if you already have a farm maybe you don't need to join the network. Or maybe you do join but don't give access to your entire farm.

As long as people look at it as a "me" proposition it goes nowhere.

Wilfrick
12-27-2011, 04:27 PM
who pays the power bills of all the people running constant renders on my computers? Me? Hmm...

if this is your problem you will not join to farm and you will not drive your render on other people's comps, you will haven't any priority, this is the advantage to have a "counter" of hours joined so the sw can assign the right class priority (it is only an idea, maybe can find a better one).
Think that there is a lot, but really a lot of people that don't turn off the comp, never... and this is all power lost, of course only idle power, less than full render, but not so little.

Think also about all the ps3 that joined to... i don't remember the name but there was a cloud computing program for a good project, all these people received nothing. Why did they made this? For the good of all.
Why I want to join to a renderfarm program without receive "money"? Because the day that I will need it I will have a really fast render and it will be a really plus for my work and the energy that I will spend to run some night of other people's render will be less than the energy I will save for my render..... think for example: your render on your comp need 20 days... with this "renderfarm" and only 50 people joined how much days you will need? less than 1 night... only few hours and you can give your work to customer... how much work more than now can you make? how much tuning can you make to your scene before push F10? How much times in past you said: "damn! it is not perfect but there isn't time to work more, need make the render because it is very slow and need a lot of days...".

Of course if you have your personal renderfarm in your office you no need all this, but how much people here have more than 1 or 2 comps?

Wilfrick
12-27-2011, 04:31 PM
Yeah, and when you are doing your renders on my computer who is paying for them?

It's not like the system is asking you to volunteer your computing power and never take advantage of the network. And if you already have a farm maybe you don't need to join the network. Or maybe you do join but don't give access to your entire farm.

As long as people look at it as a "me" proposition it goes nowhere.

yes, you said it more right than me, with less words and more effective.

OnlineRender
12-27-2011, 04:36 PM
why not just use garagefarm.net and save a wad of cash ...

it's good stuff its a positive approach and LW related , so its all good in the hood , I reckon LightWiki could take some of that strain " download speed off our server tops out at 2.5 meg and thats with my net " upload speed is hmm pushing 1 meg ! "

JonW
12-27-2011, 05:03 PM
Just looking at my electricity bill alone:

My farm, 6 computers, the whole system running is 2300 watts according to the UPS. In a 24 hour period 55.2 Kw at current rate of $0.23 plus $0.48 per day to read the meter. $13.18 of electricity & if I need to turn the air conditioning on, add up to $5.00 per day. About $18.00 a day.

Wilfrick
12-27-2011, 05:52 PM
Just looking at my electricity bill alone:

My farm, 6 computers, the whole system running is 2300 watts according to the UPS. In a 24 hour period 55.2 Kw at current rate of $0.23 plus $0.48 per day to read the meter. $13.18 of electricity & if I need to turn the air conditioning on, add up to $5.00 per day. About $18.00 a day.

None say you to give 6 comp to lightnet-farm (or how you want to call this), you can give only 1 during the lunch and dinner time... how much is it? 18$ for 6 comps x 24h.... so only 1 comp for 3 hours is 0.375$.... but with it you have access to other (maybe) 50, or 100... or... who can know now, comps.

Wilfrick
12-27-2011, 06:03 PM
why not just use garagefarm.net and save a wad of cash ... "

but garagefarm is a classic renderfarm, how I understand from their site.

Lightnet:Cloud is the "inverse" of it: there isn't a renderfarm in one place, but all computers that join to cloud become a node, so one scene can rendered in hundreds of different places in the world... the first frame at new york, the second at pechino, the third at tokyo, the fourth ad springfield on the Homer's comp, the fifth in a village on a mountain in nepal...and so on... and all frames are saved on the "central" FTP.
(of course all nodes have LW).

beverins
12-27-2011, 07:36 PM
Kts an excellent idea. I find the naysayers to be funny, because lots and lots of people already do this on facebook for the blender renderfarm. Just boot up facebook and run the app. People ask nothing in return. I run it all the time just to help out fellow artists.

If lightnetCloud could be made to run in Facebook, then that would be really sweet and solve a whole host of problems.

beverins
12-27-2011, 07:39 PM
Just looking at my electricity bill alone:

My farm, 6 computers, the whole system running is 2300 watts according to the UPS. In a 24 hour period 55.2 Kw at current rate of $0.23 plus $0.48 per day to read the meter. $13.18 of electricity & if I need to turn the air conditioning on, add up to $5.00 per day. About $18.00 a day.

Easy answer to that: Same as the blender facebook app.... Don't volunteer anything and keep as you are now.

Surrealist.
12-27-2011, 10:59 PM
Cloud computing is nothing new.

It seems to be just a very smart and astute idea to utilize resources in an efficient way.

Houdini has a cloud service app.

http://www.sidefx.com/index.php?option=com_content&task=view&id=1630


TORONTO: November 19th, 2009 - Side Effects Software, an industry leader in 3D animation and visual effects software, is pleased to announce the public beta release of new artist-friendly HQueue cloud computing tools. Studios and individuals who are looking for an artist-friendly way to harness the processing power of Amazon’s EC2 compute cloud will now be able to launch renderings and simulations from right inside Houdini.

“Studios and artists can for the first time expand their computing power quickly and securely without worrying about buying new computers or finding space for an ever-expanding farm,” says Kim Davidson, President and CEO, Side Effects Software. “In addition, students and artists using Houdini Apprentice HD can let HQueue work for them on the cloud while leaving their workstation free for day-to-day work.”

Here is a link to the explanation of the service it runs on:

http://aws.amazon.com/ec2/

I think it is something worth investigating and researching.

Now here among LightWave users, In theory, it sounds great. If I understand this correctly, I think it has some drawbacks that would have to be considered first.

Mathematically speaking, it requires an exponential number of available computers at any given time to service the number of active users.

Even if you consider the fact of down time on potential computers, I think the numbers would be staggering.

I don't think you can realistically say that there are enough LightWave nodes available on computers around the world to make this add up mathematically to be an advantage to that many users at any given time and thus it reduces the income (or resource) potential for anyone offering the severs and management to make this possible.

For each person using this service there has to be a number of potentially idle computers available at any given time to increase the speed of rendering.

For instance if one user is to increase his speed by 10, there has to be 10 idle computers available. 10 users 100 computers and so on.

Now if there are 100 computers/users on the cloud then potentially at any given time there need to be 1,000 users on the cloud to give these 100 users 10X power. However now you have also 1,000 users/computers on the cloud that want to 10 x their power making 100,000 users/computers needed to service them and so on.

However this cloud idea is based on the fact that there is some random factors in usage/down time.

So lets say if it is 50/50 then this cuts the exponent in half. But still it gets into staggering numbers.

So the only place this thing can fall down is that it requires an individual entity to set it up and manage it. It takes time to advertise it and get people to join up and incentive to use it - which is increased render resources - an exponential number mathematically. So that the more people who join means the more people who have need and the need grows exponentially based on the members who join.

This is why it works great for something like Blender and Facebook. There are probably a much much larger percentage of people with Blender installed who use it sparingly compared to the number of users who need it for serious tasks, cutting down the exponential member/need factor quite a bit.

I think this is a good idea but it requires some research into the actual numbers first and see if it can work out mathematically which is the biggest thing in question compared to the time it would take to set up and manage.

That is the way it seems to me in my simplistic realistic way of thinking. That is assuming I understand the way cloud computing works. And my understanding of it is that it works best on tasks that require mass gather ing of information or tasks that masses of people can give access to without surrendering their computers entirely for CPU time and for apps that they already have installed (such as virus protection) or apps they can install for free. This offsets the exponential factor enough that it works out mathematically on the average.

LightWave is confined to a very specific group of users in a very specialized field.

And the math here would be more of a factor in making it work than in other situations.

And this is why you'd have more people quoting rates for energy to use computers and so on. A larger majority of LighWave users are professionals or at least serious users than of something like Blender. In Blender you have a lot more users who dabble or use it very sparingly.

JonW
12-27-2011, 11:14 PM
Also some scenes can get very large 1GB or so & this needs to be sent to every node.

Surrealist.
12-28-2011, 02:11 AM
Actually if my understanding is correct, it would reside on the server and be fetched by the cloud app which is installed on the host computer.

http://www.joejustice.org/books/lightnet-cloud/how-add-computers-your-render-cloud

But there are other complications too.

Not all LightWave installs are equal. Not everyone uses the same plugins has the same versions of LightWave available and so on. And most LightWavers rely heavily on 3P plugins.

Then there are certain limitations with network rendering. For instance if memory serves me correctly, there are some procedural textures that render differently from CPU to CPU so there could be problems there.

This is not just a simple solution of having people coming along a joining in the nextwork. It would take a lot of management on a machine by machine, scene by scene basis.

You would have to attract people who were not only LightWave savy but computer savy and that will narrow down your user base right there. And those who are savy will have to have an incentive to deal with all of these technical issues - that would be render time. And so you have a larger percentage of people joining who are looking for the benefit and increasing the use of the service and thus rendering it basically redundant and useless on a purely mathematical basis.

Not that this is a bad idea. But you can see why people set up these things as services and charge money to individual users. It makes much more sense from a technical standpoint to use a render farm that is set up for LigthtWave.

If this is to work, I feel as if a lot of effort would have to go into working out the various technical pitfalls of managing something with a ton of variables.

If that effort was made and the issues resolved to make it practical and attractive, then yes, good idea.

JonW
12-28-2011, 02:50 AM
Also none of my PCs are connected to the internet & never have had any anti virus software. I don't want to go down that path. I had AVSW on one box years ago & it gave me endless problems. After cancelling, the software company's viruses quite quickly stopped coming. Which was always to my Mac anyway! Food for thought!

OnlineRender
12-28-2011, 05:33 AM
http://www.greenbutton.com/


got to admit joe justice is the coolest of names .

Wilfrick
12-28-2011, 08:06 AM
ok, all what you all wrote is right, in theory.

Pratically there aren't so much problems, because:

- how much time now is your comp busy to render? always? I think no. My comp is busy to render maybe 5 days every month. I spend the largest part of time to set models and build the scene, only at the end I make the final render, but before there are a lot of tests, meeting with customer, tuning, etc. In all this time the comp in the lunch and dinner time and in the night is turned off. So, in one month, I can "give" my comp a lot of hours to "cloud". Maybe some week no, maybe some week yes, no problem, if I can I give, if I can't I don't give, none force me to use the cloud all the time, we aren't play in WOW to must be connected. So, on 100 comp, how much hours in one day are busy and how much are free for the cloud? I think more than 10 comp free every 1 busy, but also if more busy I don't see this problem.

- Of course need responsability from us: we must send only the final renders or the very big tests.... In my work I make a lor of test animation renders, but they are really fast, only to show something to customer and after need change, tuning, etc. If someone send these test to cloud the render will be more slow than make himself.
Think now to the very very hard renders, with glass, GI, reflections, refractions, metals, fullHD, big AA, motion blur, DOF, and so on.... every frame can take 1 hour, 2 hours... these are the renders to send to cloud. I really don't think that you make also the tests (that you know you will need to modify) with this rendertime... Tests must be very fast, of course not for cloud, but the final render yes.

- Responsability.... don't send 1GB scenes! Maybe your scene is 1GB but need only 2 minutes to render, and maybe you have an other scene of 30 MB that require 2 hours every frame... You will send what you know that can be sent... we aren't children, we are all "professionist", we know comps, ram, connections, limits..... if not the cloud isn't possible.

- Like I wrote in first post, of course can rendered only scenes that require standard plugs and the other will go only on comps that have the required plugs... Need some changed to lightnet:cloud software, like checks of plugs, lw version, ram, priority....

- It isn't expensive, need only 1 good FTP. The more hard thing to do is the changed to software to can menage all users. If users aren't egoist the game can works, if someone want to receive all and give nothing the game finish before begin.

This isn't a new lw function, it is not something that you must have... it is only a plus, someone can help and receive, someone no need, someone can't. None must feel pushed, it is only a good opportyunity, nothing more. Who see problems can don't join, no problem for other, very simple. If someone have more good solution or suggestion can give his idea to the project and it will be more good.
Forum users can be a good test, no need for now million of comps, all can be good with less than you can think... maybe... I hope.... or maybe I take a lot of sun and I'm crazy, very simple this :(

dsol
12-28-2011, 08:27 AM
I think a totally altruistic and honor-based system - "for the greater good" - is sadly not going to be workable. It only takes a tiny percentage of people to abuse the system for it to fall on its ears.

So instead I'd suggest an algorithmic solution. One existing solution to the issue of "fairness" when it comes to distributed computing is to implement a Ratio system, like BitTorrent. So people who simply use the net to render their own stuff and don't supply any CPU time in return get a lower ratio. And if they go below a certain threshold, their access to the net is restricted or revoked.

This doesn't get around the other big issue mentioned elsewhere: of large scenes taking long to download. However, this could again be tackled by smart software design. So....

LegionRender™ (hehe)


Open source client. Supports P2P distributed computing tasks with high levels of security and anonymity - both for active users and BG cloud nodes.


Support for automated download and re-use of large scene files - designed with particular emphasis on 3D rendering. All downloaded content is encrypted, hashed and run in a sandbox - source files and renders are not viewable by the machine rendering. All source file downloading is done via an integrated and embedded BitTorrent client. The scenes are archived with all content into an encrypted .LegionRender file. This format builds upon the BitTorrent format by supporting an extra layer of built-in security. Only the original creator of the file can unlock the content - other clients have to handshake and share the key temporarily. Rendered frames are encrypted and stored in a temp directory on the client machine, until they are successfully uploaded to the scene creator's client (which saves them to a directory they specified)


Automated Load-balancing. The Tracker and clients work together to maximise efficiency - and fairness! A "render Ratio" score is kept for each client. If that user wants to render a very large scene, then they need to be very aware of how that will affect their ratio. If they're smart they'll have built up their ratio to very high levels in advance. Kind of like farming :)


Third Parties can easily wrap their standalone renderers as plugins, to be installed and used by LR clients. Exporters can be written to export directly from your 3D software to a native .LEGIONRENDER file, ready to be run on the network. EDIT: Here's an even cooler idea - how about if the actual renderer was actually bundled into the .LegionRender file? That way you don't even need to worry about the end user having the right set of plugins - or the most up to date versions!


Open Source Tracker. Server-side admin software. Run your own trackers - great for large and secretive VFX firms that don't trust the inbuilt security features of LR and want to keep all data (even encrypted) inside their own network. The tracker also supports client management to monitor end users and to control individuals who abuse their privileges. Fully automated responses - you can set multiple levels of email responses and actions taken, for different levels of Render Ratio.


Well, I can dream, right? :)

Wilfrick
12-28-2011, 09:40 AM
I think a .....

oh... wow!
more simple talk, more good english (of course, you are from London!), more precise ideas.
Better than you explained can't do.


p.s. ...... LegionRender? hey hey hey, how much I like this name!!!

OnlineRender
12-28-2011, 09:49 AM
Onlinerender is better (",)

dsol
12-28-2011, 10:08 AM
Now I just need to find some coders who've worked with Bittorrent, Sandboxing, Encryption and (ideally) virtualisation. Since my own coding skills are pretty basic (a bit of C, Obj-C and Java only)

OnlineRender
12-28-2011, 10:47 AM
Now I just need to find some coders who've worked with Bittorrent, Sandboxing, Encryption and (ideally) virtualisation. Since my own coding skills are pretty basic (a bit of C, Obj-C and Java only)

http://www.utorrent.com/community/developers

Surrealist.
12-28-2011, 05:55 PM
ok, all what you all wrote is right, in theory.

Pratically there aren't so much problems, because:

- how much time now is your comp busy to render? always? I think no. My comp is busy to render maybe 5 days every month. I spend the largest part of time to set models and build the scene, only at the end I make the final render, but before there are a lot of tests, meeting with customer, tuning, etc. In all this time the comp in the lunch and dinner time and in the night is turned off. So, in one month, I can "give" my comp a lot of hours to "cloud". Maybe some week no, maybe some week yes, no problem, if I can I give, if I can't I don't give, none force me to use the cloud all the time, we aren't play in WOW to must be connected. So, on 100 comp, how much hours in one day are busy and how much are free for the cloud? I think more than 10 comp free every 1 busy, but also if more busy I don't see this problem.

- Of course need responsability from us: we must send only the final renders or the very big tests.... In my work I make a lor of test animation renders, but they are really fast, only to show something to customer and after need change, tuning, etc. If someone send these test to cloud the render will be more slow than make himself.
Think now to the very very hard renders, with glass, GI, reflections, refractions, metals, fullHD, big AA, motion blur, DOF, and so on.... every frame can take 1 hour, 2 hours... these are the renders to send to cloud. I really don't think that you make also the tests (that you know you will need to modify) with this rendertime... Tests must be very fast, of course not for cloud, but the final render yes.

- Responsability.... don't send 1GB scenes! Maybe your scene is 1GB but need only 2 minutes to render, and maybe you have an other scene of 30 MB that require 2 hours every frame... You will send what you know that can be sent... we aren't children, we are all "professionist", we know comps, ram, connections, limits..... if not the cloud isn't possible.

- Like I wrote in first post, of course can rendered only scenes that require standard plugs and the other will go only on comps that have the required plugs... Need some changed to lightnet:cloud software, like checks of plugs, lw version, ram, priority....

- It isn't expensive, need only 1 good FTP. The more hard thing to do is the changed to software to can menage all users. If users aren't egoist the game can works, if someone want to receive all and give nothing the game finish before begin.

This isn't a new lw function, it is not something that you must have... it is only a plus, someone can help and receive, someone no need, someone can't. None must feel pushed, it is only a good opportyunity, nothing more. Who see problems can don't join, no problem for other, very simple. If someone have more good solution or suggestion can give his idea to the project and it will be more good.
Forum users can be a good test, no need for now million of comps, all can be good with less than you can think... maybe... I hope.... or maybe I take a lot of sun and I'm crazy, very simple this :(

lol!

Well this sh*t is all way over my head. Just my simpleton way of looking at it.

Seems exponential (constantly multiplying by itself) math to me. :)