Script for distributed rendering ?

[bernhard]

Hi there,

is it possible to get a script that allows to render with the Standalone Version and using distributed rendering ?
I do have about 100 multi-core machines.

Needed for Linux Fedora, 64bit with Vray Standalone 1.5 version 28-02-2011.

Any little help highly appreciated !

thx and till later,
Bernhard

[ultrasonic]

what exactly are you looking for? a rendermanager for animation or just one command for a single frame?

[bernhard]

well.. actually I am having troubles with the 100 multi-core and the distributed rendering... it works fine when I use just 20, but when I use 100 machines... I can't get as many machines rendering.
so I was wondering, if there is a limitation, or something is broken in the distributed rendering script we use.

To double and triple check, I use a very small bucket size, and go with a 2k render ... but for some reason... I am limited with the machines

that's why I was thinking, maybe something is limited in the distributed render manager I use.

thx for any little hint !

[pechart]

I think i remember a thread where a guy had issues with rendering on more than 60 cores per machine.

Also, do you have enough licenses? One GUI license includes 10 DR lics, as far as I know.

[bernhard]

thx a lot.. yeah.. that shouldn't be a problem....

[Zdravko]

V-Ray single license has limitation for DR to use up to 10 render machines. If you use more machines to render with V-Ray you can contact Chaos Group sales team to offer you appropriate solution for render farms – it could fit better your needs.

According to DR rendering question you must have at least these flags “-distributed=1”, -sceneFile=”path/to/vrscene” , -renderhost=1.1.1.1, -portNumber=20207 in the command line to run distributed rendering with V-Ray Standalone .

[bernhard]

well.. I do have 15 license just on the standalone... which does make 150 machines...
will need to check our dr manager....

[Zach Gray]

Bernhard,

I wasn't getting very good performance using DR, so I switched to Qube and .vrscenes with machines individual frames. 10 computers working on 10 frames seemed to be much faster than 10 working on 1. I'd be interested in hearing from Chaos on this one, in case I could get better performance somehow.

I'd be happy to share my Qube simplecmd template for .vrscenes if you end up going in that direction.

Zach

[ALFRED_ST_MEDIA]

Zach,

I would be interested in hearing what the Chaos team has to say, but I would imagine your bottle-neck in rendering one frame on 10 machines vs. 1 frame per machine would be mostly due to network latency of splitting up a frame, then re-assembling it, as well as the re-assembly process itself being an overhead on a per-frame basis.

We only split single frames across multiple machines when there is a memory issue, such as 8k stereo frames.

-ctj

[Zdravko]

If you have all the machines licensed and included in one DR job there is no reason for any of them to be excluded from the job. From the other hand there is no point to use DR with more machines if the scene is not heavy enough as the network load could slow down the rendering time.
The problem could be caused by many reasons: the network capacity, the scene itself or the render manager that you are using and its settings, either.

[Zach Gray]

Zdravko,

Is there a way to update the DR interface to allow for a 'pool' of available machines that several users can utilize. For example, I have 10 machines with the slave running. Say the user is capped with 3 DR clients, and uses whichever are available. So basically when a machine is busy, it flags itself as busy. If it's available, it can accept DR connections.

We're in a volitale environment where machines can join as DR as the user logs off, but it takes a long time for that host to be considered unreachable during a render.

I can elaborate more if needed, but it's essentially a pool of DR slaves, dynamically assigned to multiple users.