VRay Renders are Slow thru Deadline

[wheely]

Hello,

As suggested by Mike Owen, I’ve started a new thread concerning our troubles with slow farm renders vs. quick local ones. The original thread is here. I will again upload my log file in this post and respond to Mike’s queries.

We are using Deadline 7.2.4.0 R on a Windows system.

1. This needs fixing -

C:/Program Files/Autodesk/Maya2015/scripts/others/dynRunupForBatchRender.mel line 33

Could you describe the fix needed? I’m guessing this fix also needs to be deployed onto all render nodes?

2. 12 x mins of loading bitmaps. - when multiple slaves on the farm try to run the Maya job at the same time, is your network file server struggling to deliver the performance? If you run the job again BUT with only 1 Slave on the farm rendering this job, do the load bitmap times become reasonable? Bear in mind, if you already have the Maya scene file open, then the bitmaps could be already loaded…

For the example in the uploaded log, the job was a single frame (1-frame) only. Thus, just one machine rendered the task. At the time of my testing, there were only a couple of other jobs on the farm, granted, about 20 machines were being used to render them. I will speak with the IT fellows here to glean more about our network setup.

4. Something that looks odd is the final output filename with a “space” character at the beginning of each fileName, like this:

" multimatte_a.0251.png" <–whitespace character at beginning of fileName
" multimatte_b.0251.png" <–whitespace character at beginning of fileName
" multimatte_c.0251.png" <–whitespace character at beginning of fileName
" multimatte_d.0251.png" <–whitespace character at beginning of fileName
" rgba.0251.png" <–whitespace character at beginning of fileName

I will post below screengrabs of our filename tokenization and the Render Elements tab of the Vray render.

MayaJobLog.txt (194 KB)

[wheely]

As I was investigating this more, I’ve noted that some of our staff PCs are rendering quite a bit faster through Deadline than our blades are. This is not the case for every single PC, but we are noticing a trend that there may be a problem with how our blades are spec’d or set up. The IT staff here is looking into comparing the fast machines with the slow ones. At this point, it feels that our issues may not be related to Deadline, but if you see anything in the logs where we can still optimize, please let me know.

Thank you,

PS- just to let you know, I’ve been getting render times between 6 and 8-minutes on the fast PCs versus 27-minutes on the blades.

[kdronez]

Hi,
this is probably not the case,but we had similar problem with one node.It was 2 times slower than the others and it turned out that its power management was set to safe mode in the bios xD.In the monitor it was showing that the CPU was utilised 100% (which was right since it was getting half the power),but was still slower xD.

[eamsler]

I’ve also seen thermal throttling be an issue. I do not think that’s the case here. Also, have you tried running the Slave on a workstation to see how that goes?

I can show you how to get an exactly reproducible render in MayaBatch outside of Deadline. I don’t believe this was a feature in 7.2 but there should be an option to include the Melscript in the job report.

This would let you start up Maya and paste that log (after some cleanup) into the script listener and check to see the performance. With the dirmap mode enabled, that would also load the scene fresh from the Melscript and get you closer to what I’d expect from Deadline.