Octane Network Render Problem/Question Tweet

[Karlifetz]

So, I recently bought Octane Enterprise + a slave Subscription because I wanted to turn my old rig into a slave to speed up the overall process.
Octane Network Render in itself works fine on still images with rather huge sample sizes. Everything is detected, when it actually kicks in, the GPUs on my old rig display Dedicated GPU memory usage.
I always thought that if I had a minute of rendertime before Network Rendering, maybe I could shave those 60 seconds down by quite a bit and render my needed sequences way faster.

So I sent a scene and the issue is that Network Render needs to send over between 350-450 MB, sometimes more. Generally that would be fine, if it was a huge sample size for a still image.
Before that transfer even visibly begins, the scene is already done with rendering, the amount is bigger on Frame 0 and then stays the same after Frame 1 without any progress over the coming frames.
Does that mean I actually have to pick sample rates that give the data enough time to transfer? If that is the case, I probably misbought. I just couldn't find anything about it in the Documentation, same with the "Use Steal" under Network Render Settings, I don't know what that is. Maybe I overlooked it, I'm sorry if I did but I'd appreciate it if anyone could help me out.

As for the example;

This would be on Frame 0.


And this is what it would be like a couple frames down the line, nothing changes.


When I crank this specific scene to like 2500 samples, almost the same is happening.
- Renders in like 2:30 minutes without any progress being made on the transfer on Frame 0 (Around 440MB)
- Then on Frame 01 it suddenly starts to transfer a total of 2095MB after being on 346/346MB for the first few seonds. Within the time of rendering, it was able to get to 1100MB out of 2000MB, then it goes on to...
- Frame 02... back to 346/346Mb again. Within Frame 02, it was able to finish those 346, then it goes dark green saying: NetRender Gpu: 0/0 Slaves: 0, turns back red and starts the above mentioned download of 2000MB again.
- Only to completely start over when it reaches Frame 03.

And this goes on.

I am seeing a 100% Network Usage and a rate of 98Mbps of Recieving Data in the Task Manager on the Slave Machine.
I am really baffled, is there any way to send all of the data at once before rendering begins?

Edit: I restarted the client and tried again with the 2500 samples scene. This time it worked, it started to transfer starting at Frame 01. When I started it again and tried with 800 samples, it does take about 1m-1m20s for each frame to complete but it wont even start the transfer. I restarted it with 2500 samples and this time, it's the same as above - each transfer is completed after about 40 seconds starting at Frame 01 Network on the slave machine is always at 100%, recieving around ~95mbps. Which I don't know why it wouldn't start doing that with the 800 samples test. So the same problem persists, I probably won't really get a speed increase unless the sample size is big enough to let each frame transfer all of its data - am I assuming this correctly?

Thanks in advance.

[bepeg4d]

Hi,
which exact version of plugin are you using?

It would be nice to have the c4doctanelog.txt file from the Master, and the daemon log from the Slave.

To do so, before starting to render in Network, please, copy paste the attached `octane_daemon_log.txt` file in the Slave 2020.1.3 directory:
octane_daemon_log_flags.txt.zip

Then open a scene in the Master, connect it to the Slave, then go to c4doctane Settings/Others tab, and, at the bottom of the panel, enable all the Log option checkboxes:


Then, load the scene in Live View, then in Picture Viewer, or until crash, and share the c4doctanelog.txt file that you find in c4doctane directory, or in the path specified in Settings/Others panel, thanks.

Then, from the Slave, navigate to `C:\Users\UserName\AppData\Local|temp`, or write %TEMP% in the Explore path field.
Locate the octane_damon_log.txt file and share it to me, thanks.

ciao Beppe

[carpo101]

We're having exactly the same issue with the data transfer time making it not worthwhile to use network rendering.

Is there any workaround or fixes for this?
Cheers,
Tom

[bepeg4d]

Hi,
maybe Out-of-core is active in Workstation and/or Render-Nodes, with that scene.
We need to see some logs.

Please, open the scene, then go to c4doctane Settings/Other tab, and, at the bottom of the panel, enable all the Log option checkboxes.
Then, load the scene in Live View, then in Picture Viewer, or until crash, then reopen c4d, navigate to c4doctane Settings/Other tab, press the Generate log report button, and share the octanelog_report.zip file, thanks.


ciao Beppe

[carpo101]

Hey Bepe,

We've saved a log file. I think the action starts after the save at:

[2021-01-11 14:25:57][0] save

From there we started a Net Render using one other machine on the network.

The Master has 2x3090s, the slave has a 2080Ti + 1080Ti.

We found that on this occasion there was an improvement in the render times (approx 4min 30s pf compared to 5min 30s pf).

There was approx 2.5Gb of data transfering on each frame and was able to trasfer before the frame had finished rendering. The issues have happened for us when the amount of data transfer on each frame is higher (therfore takes longer to transfer) and the frames finish faster (say lower sample count), not leaving enough time for the slave to contribute.

Do you need us to get a log of the slave too? I wasnt sure where to install the log file and also, we're on the latest version of Octane C4D so we may need an updated file?

Thanks in advance - you've helped us out loads before so appreciate it greatly!
Tom

[bepeg4d]

Hi,
there are several c4d Layer shader in the scene, that is not supported, especially in Net render.

Till 2020.1.5, you can replace it with OSL color_layers.osl node that you can find in LiveDB/OTOY/Procedural Textures/Utilities category:


But since you are working with Ampere GPUs, my suggestion is to update both Workstation and Render-nodes to 2020.2-RC6.
In this way you can use the new Octane Composite Texture node to replace the c4d Layer shader.

Please avoid to use c4d Shaders in general, especially with Net render, unless they are c4d Noise or Gradient shaders, that can be automatically converted to OSL, with the proper option in Settings/C4D Shaders tab:


ciao Beppe

[carpo101]

Bepe - you are a legend, thanks - we'll let you know how we get on