Thoughts for Render farm?

[GustavTheMushroom]

Hi Ono,
I just had some thoughts on the render farming stuff in indigo.
Could you build it into the gui? I mean for instance, you cold connect all the render "nodes" before you start the render, and then tell it which file to render and it will transfer the files (In Sc Files including textures etc) to all of the render nodes and they will start up once they have received the files?

Seems like it would be easier to start renders like this, you could run indigo on startup so all you'd need to do is to start a machine up and it'd be ready to go. You shouldnt have to tell it to do anything because all the data will be transfered. Indigo could just be permanently listening on some port so it could detect a render....


Please tell me this makes sense lol. Im not too good with telling other people about my ideas....

Oh I forgot to mention how I thought the "connecting" would work... you'd just have to tell the "main" indigo what machines the renderers are and because indigo will already be listening it shouldnt be an issue should it? some sort of fail safe would need to be there in case not all of the render machines have been switched on etc etc.

I realise that what im talking about works in complete reverse to how it currently works.... Is there a reason it works the way it does at the moment?

To me this sounds complicated, granted, but i think in the long term it would be good to have a system like this. It just takes so long if you constantly have to go to each machine and start a renderer and transfer the files yourself etc etc etc...

Just some other ideas within this one:

*Some sort of version checking on the files (ie a checksum) to see if the files on the both central and node indigo are the same.

*A transfer button? To transfer all the files so you could control the start of the renders better?

*A quick transfer button to refresh only the files that need to be (ie those with different versions)

*Some sort of monitoring panel (CPU, memory etc of all the nodes)

*Individual node displays to look for differences?

I know my ideas are reasonably big.... and annoying XD.
But I hope my ideas come in useful...

[zsouthboy]

More than one of us forum members are working on programs/scripts to do exactly that

[Deus]

Let the rant begin.

How these systems work and the things you point out are fucking obvious (excuse my french)

This is what I would like (joke but funny)

* I would want to press as many buttons as possible for 7 minutes to start a render
* I would like to not have version checking between hosts and client so stuff can break after 3 days of rendering
* I would like to have to transfer my scenes to the nodes by tape and then unpack them manually.

lol

damnit guys

[zsouthboy]

Here's how my program will work:

Start something on the "server"

All clients automatically start that particular scene.

Kill indigo on the "server" and the clients automatically close.


note that file transfer, etc. is a pain in the ass and needlessly takes up network resources.

Therefore my farm simply has drive "Z" mapped to the shared indigo folder on the main server. Then there is no worry regarding file versions, textures, etc.

As a plus, the tree_cache can be used by all clients, rather than they all compute meshes themselves.

Right now on my farm I'm simply using RDP to connect to each client and starting a batch scripted indigo manually.

[GustavTheMushroom]

Lol.

Im an idiot.... I shoulda thought of the drive thing....

Makes things a lot easier. *slaps self*

Could it work on a shared folder rather than a mapped drive? Just wondering...

[mrCarnivore]

@zsouthboy: Nice script. When will it be finished?