I read about the load balance and how it should get better. For some reason, mine is getting worse. My laptop, which is the slowest of the machines seem to be getting all the work while my Xeon computer, the fastest, gets the smallest load. (As seen in the picture where the fastest machines after a couple of renders are down to a sliver)
Looking at that image and reading what James said, I think this must be an issue of the specific files being rendered. I'm not sure CrowdRender knows what pieces of your render are more complicated. this could maybe be done at some level with a fast low quality render and some data points beforehand?, but that's a pain.
I just rendered a test animation that doesn't have much of anything going on for like the first 50 frames in the left half. It's pretty much a solid colored plane. What happened over 8 computers rendering this was 5 would render really face, the other 3 take twice as long. Then the load balancer I believe would calculate and swap out the slow ones for the next frame, rinse repeat. Once the whole scene had as much going on everything balanced very nicely.
Hi Rocks, you're correct about the cores. The current version is using the number of cores as the first estimate since it is likely that more cores equals more speed. However, we're changing this in 0.1.6 since this assumption doesn't always hold, especially where GPUs are concerned!
0.1.6 is the next release we're working on, it will have some fixes for the load balancer to help with poor balancing. We're also still crowdfunding for funds to rebuild the integration with blender's render engines and fix the issues with computers not restarting or continuing to render if there is a sync issue.
This is the final render as the hardest spots as you can see from all the pic ended up being done by the laptop. I am guessing after looking at things, the laptop was working harder cause of the 8 cores and I believe the others have 4 core CPUs each. later on it seemed to balance again but goes back to still giving the laptop more to work on then the others. An other thing I seen happen once, if a computer unsyncs, it will not restart its job. One thing I was going to suggest was maybe having a way to restart the job or have an other computer take over the failed or incomplete job.
Hi Rocks, can you let me know what the scene looks like that you are working with? The scene does have an impact on how the load balancer works (if you check www.crowd-render.com/crowdfunding out you'll see we're currently trying to raise some development cash to fix this).
If you're working under an NDA, perhaps you could test a different scene and see if you get the same issue. Alternatively you can try deleting your config file for crowdrender and seeing if this fixes the issue, the config file stores the data for the load balancer. Resetting that data by removing it might help.
Generally, we can fix stuff like this if we can reproduce the same issue ourselves, so if you can share the blend file and a bit more about your hardware we can try making the same thing happen here and if its a bug we can fix it. You're of course welcome to try the method above for resetting the LB data.
If you don't want your files in the public domain, just e-mail us at info at crowdrender dot com dot au and we can setup a file transfer for you to send them directly.
Looking forward to hearing from you soon :)
Xeon finishes first but only a very small section is done with it.