Slow network rendering of multiple railclone objects on windows 10

[BB3977]

I've noticed when rendering multiple railclone pro object in a scene I get slow render times on the main licensed workstation but normal render speed on the render nodes. This can be more than double the render time it takes to render the same frame on the render node as it does on the workstation with the railclone license server. A scene with one railclone object doesn't make much of a difference in render time but add more than one you start to notice a difference.

Workstation (Windows 10pro 64bit)
Overclocked i7 4.5Ghz
32Gb ram
8x Samsung 850pro raid 0 data drive shared over network
Intel 750 OS drive
3ds max 2016 sp1
Vray 3.20.03
Railclone pro 2.5 gui license

Render Node (Windows 7 64bit)
Overclocked i7 4.5Ghz
32Gb ram
Samsung 850pro OS drive
3ds max 2016 sp1
Vray 3.20.03 render node
Railclone pro 2.5 render node installation

[iToo]

At first glance, there is nothing in the license system that may cause that problem.

What network render mode do you use, Backburner, DR or other ?
Have you tried rendering in the workstation but configured as slave node ?

If not, please try it and compare render times. In this case the license is not used, so the test would give us some clue.

[BB3977]

Render farm setup:
3ds max 2016 sp1
Backburner 2016
Backburner Manager running on dedicated server
Workstation “Pivotal_WS” Windows 10 pro 64bit i7 4930K 4.5Ghz
Node 06: i7  3930K 4.5Ghz
Node 07: i7 5820K 4.4Ghz
Node 08: i7 5820K 4.5Ghz
Node 09: i7 5820K 4.5Ghz
Node 10: i7 5820K 4.5Ghz


Test scene 01:  Static scene no animation, no GI with local maps (Attached)
3 railclone objects attached to 3 individual splines, 1 vray dome light and 1 vray plane
Test scene 02:  Static scene no animation, no GI with local maps (Use scene 1 and collapse all railclone objects to meshes as it generates a 700mb file)
3 railclone objects collapsed to 3 meshes, 1 vray dome light and 1 vray plane

Tested workstation with Railclone installed as slave mode makes no difference to render times so it's not pointing to the license server as the issue.

Rendered directly through 3ds max 2016
Test scene 01 :Workstation “Pivotal_WS”  only 2min 10sec
Test scene 02 :Workstation “Pivotal_WS”  only 1min 34sec

Rendered through backburner
Test scene 01 : Network rendering see attached “test scene 01”
I started off with 2 render nodes and the workstation for the first few frames and then added 3 more render nodes after frame 11. As you can see from the times in the manager the render times went up.

Test scene 02 : Network rendering see attached “test scene 02”
As you can see from the times in the manager in test 02 the scene renders much faster, as I would imagine that the meshes are cached in memory at the start. The workstation is also on par with the other render nodes too.

Could this be more of an issue with windows 10, 3ds max 2016 or backburner 2016 as I don't remember having this issue when running railclone pro on windows 8.1pro, 3ds max 2015 and backburner 2015.

[iToo]

Hi, thanks for the detailed tests. I've repeated them here:

i7 980x 3.8 Ghz
Windows 7
Max 2016
RailClone Pro 2.5.0
V-Ray 3.20.03

Test01. Using RC:



Test02. Collapsed to mesh:



As you can see there is difference, but not so much as in your case. Unfortunately we have not a similar computer with Windows 10, so i cannot test that configuration.

Instead, i tried using the latest V-Ray nightly build (3.25.01) with a custom RC compiled with that version. There is a significative improvement:

Test03: Using RC:



Test03: Collapsed to Mesh:



RailClone rendering is even a bit faster that using a pure mesh.

Sincerely, i have not idea what may be causing that problem in your workstation. It may be due to RailClone, V-Ray, Max 2016, Windows 10 or a combination of all (most probably).
But in this case i suggest we wait that V-Ray 3.25 is released (we'll publish the RC update short time after) and try again. Hopefully the situation may change to better.

If you have access to the V-Ray nightly builds and want to try it, i can send you the RC update.

[BB3977]

Thank you for looking into that for me. Vray 3.25 looks very promising then!

One thing I did notice was the more machines I added to the render manager the slower the windows 10 machine became.

I'll have another play around with this over the next few days and see if I can pin it down to anything in particular. I have a spare copy of windows 10 so I can try that first on one of the render nodes and see if it has the same issue.

I'll report back with the results.

If anyone else has a similar setup it would be good to know there experience too.

[BB3977]

Ok,

Just installed windows 10 on my render Node-10 and completed a test render on the network. Looks like windows 10 could be the issue after all. As you can see from the attachment, Node-10 and Pivotal_WS (windows 10 pro) are considerably slower than the other render nodes which are operating on window 7. This could still be an issue with 3ds max 2016 as it's not fully supported on windows 10, so this may not be fixed until 2017!

So it looks like my options:
1. Collapse my railclone objects to meshes before sending to the network and wait for a fix
2. Downgrade to windows 7 on all machines....and wait for a fix!

Is there anyway you could test this on a windows 10 machine to see if there is a an issue with railclone and windows 10?

[iToo]

Thanks for the information. It's really a strange issue.

We have a Windows 10 machine but its hardware is different (slower), so we have not a reference to compare render times.
Anyway, we'll test it under several conditions and post the results. I hope we can get some clue.

[BB3977]

Thanks Carlos,

If you need me to test anything at this end let me know.

[IT1371]

I suggest you turn off the SUPERFETCH service on all your Windows 10 machines.
Already posted about it in Chaos Group forums.

[BB3977]

Just tried that and still no difference. Superfetch only turns on if you're using hard drives (Spinning disc) and not with ssd's or Raid arrays anyway, so it shouldn't cause an issue.

[iToo]

Thanks for the advice. I tried it as well, but unfortunately Superfetch has no effect in this case.

Actually we can reproduce the problem, but don't know what cause it. We continue investigating.