Rendering jobs with Softimage and Deadline

[sirdavid32]

Hi, I recently wrote to the Thinkbox software: Deadline to get a free evaluation from them.
I got the software download link, and installed the Repostry on my "server" machine, and the
executable apps (monitor, launcher, slave, etc..) on the slaves.

I activate my second slave "B_Machiner" (I´m on free version with 2 or less slaves), and on the monitor (I activate the monitor from A_Machine, to put my B_Machine available as slave for the render).

My softimage project targets the output render to itself through network path : //Machine_A/xsi_project/Render_Pictures so my B_Machine could have access to the render. I also shared the folder so it can
be accessed through the network.

But my Machine_B keeps thinking the render should be locally (as the local path of my A_Machine) and thus
generates ERROR 200 - file location not available.

I´ve read the manual and it mentions all locations should be available not as relative but as complete path location, which is what I´ve checked through the entire process.
I might be overlooking somewhere, but I have no previous experience with Deadline.

Could someone lead me into the right process of creating jobs and sending them through a network of 2 machines? Please?
Thanks.

David.

[everyframe]

Hi, I'm somewhat familiar with deadline. But I'm not quite understanding the problem your having. How are you submitting your renders? From within deadline, or through SI with the submission script?

I'm fairly certain you need to map the network drive. So that the shared machine a drive is seen as a local drive for machine b.

This is what I had to do in Linux to get deadline working (NFS Share), as it would not utilize a shared drive through nautilus.

The reason being in Linux, is that a "shared" network drive is only available to nautilus, not deadline. Nautilus is the file browser. Like windows explorer. So even though your able to read/write to machineA from machineB - deadline is unable to.

for Instance, on my render nodes. The network drive they render to and read files from appears as an internal drive, and is the same throughout each machine, even though it's a network drive. Every machine sees it as /media/OrvilleRedenbacher/Bagheera/Projects/.

Also, make sure in the deadline settings, that the path to your executables for softimage are correct. They should be for windows, but for linux, I had to change them.

are you using two machines and a server, or is machineA serving/hosting the repository?

[sirdavid32]

This is the workflow I followed:

1. Installed Repository on "master" machine
2. Installed .exe on the "slave" machines.
3. Started softimage, and created a project not as local, but in a "mapped" drive letter (W:/)
The mapped drive, pointed to a local shared folder, so all network will have access to it
in the format: W:/XSI_Project/
4. From there on, all the paths were configured with the mapped drive and thus I work on my
project. Do texturing, lighting, etc. Time to render...
5. Send project to deadline, and activate slave machines (2). 1st slave is the origin machine, and the 2nd
one another machine on the network. with USE XSI BATCH option activated.
6. I open a Deadline Monitor on the Original Project Machine (Slave1). And effectively I´m watching
my frames render swiftly!! FINALLY!

Thank you very much for your help!!

Now I would like to know how many frames I can assign to each slave. Say I´d like to do a 100 frame
animation (test) 20 - 20 each.
I will monkey around some more, but any additional tip, would be so helpful!

[everyframe]

Sweeeet. Glad you got it figured out. Tech support at Thinkbox is super helpful too. Even with just a demo license, they try to help out as much as the can. within reason obviously.

In my experience, managing a render farm is more of an artform than a science. Achieving fast, efficient renders won't happen automatically. It varies depending on which softwarepack your rendering with, but It will more or less be about finding a sweet spot between task size, and concurrent tasks for the amount of cores in a slave, ram, and how expensive the scene is to render, Enourmous cache files being loaded, etc, etc, etc.

It gets even more complicated when you start introducing multiple slaves with different cpu/ram configurations, and setting up the appropriate pools so that a balance of how many jobs rendering to how fast those jobs are rendering is optimal.

Create a test scene. and just keep resubmitting/rendering it with different task size, and concurrent tasks. Things can speed up 200% with the right settings - versus 1 frame per task, and 1 task per slave. With houdini, It almost doesn't matter. It's gonna render the tasks with full processing power. But i've noticed things to be more finicky with maya or Soft, and requires constantly dialing in these setting for optimal performance almost on a scene by scene basis. anyone else?

You also wanna avoid having a slave checkout lots of frames, when there are other available slaves sitting around idle. that typically happens towards the end of a sequence though, and when there are faster slave machines in a pool with slower slave machines. Since the faster slaves will finish their tasks first, they will sit around idle while the slower slaves finish up the tasks they checked out to render. I hope that makes sense.

Like I said, It's an artform in itself. I've worked for studios that don't give a shit, and utilize maybe 10% of the actual processing power of their farm. Just cause they can't figure out setting up appropriate pools, task sizes, and concurrent tasks (some machine have 24 cores). then doing some speed tests, and most importantly - training their employees to properly use/submit renders to a farm, not just hail mary frames at it.

Sorry for the exhaustive reply ;)

Goodluck!

[sirdavid32]

Thank you for your tips. I totally get it. The slow/fast machines on the same pool.
I will keep an eye as I struggle through the parameters you mention, because it makes sense: more cores,
more idle time somewhere else.

I was trying to setup deadline 5.2 to 3DsMax 2012. I did set it up on Nuke and Softimage. So far, pretty nifty.
But the Script folder of deadline has a .py script for max, and I did downloaded the blur plugin to read .py files,
and installed python. 2.2.6 for X64 windows, but so far, the "submit to DeadLine" fails to load.

So I loaded DeadLine Monitor to upload the 3DsMax file from a mapped letter drive and assigned the frames.

When I start up slaves, only "origin" machine renders the file while the other slave says there´s a problem with 3DsMax
(the machine where I installed Python and blur tools). On both machines I have installed Vray
but right now I´m testing it with a "normal" renderer.

Hope this gets clear.
Thanks.

David.