Maya Mental Ray Network Render - SLOW


#1

Hey,

I’ve been setting up a small Maya Mental Ray network render farm and after weeks of testing, I seem to come to the conclusion that network rendering as actually SLOWER than rendering on the local machine!
Is there anyone that might be able to shed light on this? It can’t be right as many studios have to use distributed rendering in order to pump out frames at a reasonable production time.

Here are my Software and hardware:

SOFT:
Windows 7
Maya 2014
Mental Ray 3.11.1.13
Shave

Slaves running Mental Ray Satellite 3.11.1

HARDWARE:
Master: PC, Xeon E5-2650v2 2.6GHz, 16 cores / 32cores hyper threading, SSD
Slaves x 4: Power Mac G5 running Bootcamp Win 7, Xeon E5462 2.8 GHz 8 cores, SSD

As my LAN only makes 0.65 Gbps, I copied all textures to the local machines in order to reduce network overhead. Still, the network render is slower than on the local machine.
Would using a dedicated render farm management software make things faster? Any suggestions would be highly appreciated.

Thanks


#2

I’ve just set up a master slave system with a windows pc and a mac mini running satellite (2014). I’m getting a speed up of almost 2x, so no you should be able to see a speed up. Have you checked that maya is loading the maya.rayhosts file correctly, you should be able to see in the output window that this is working. The output window should also output progress to show which hosts are being used.

Try a batch render, and compare selecting “Render on the local machine” only, against “Render on network machines only” in the batch render options. That will tell you what performance you are getting out of the render farm. The power macs are quite old machines so I don’t know how well they will do, but it certainly shouldn’t be slower.

You could also try pinging the slaves from the master machine to make sure there aren’t any network issues.


#3

If your frames are quite fast to render (5-10 minutes) you will hardly see any improvement with distribute rendering. Satellite rendering is slower at the beggining because of the network traffic increase. Master machine has to send all the data to the slaves. If your render farm is composed by a lot of slow machines the network overhead will probably kill all the benefits of the distributed rendering. I usually render with satellite hi-res stills that require some hours to be calculated and the speed gain is nearly linear. If you have to render an animation it’s a lot better to make each machine render a single frame at a time, in my opinion.


#4

Thanks for your replies guys. Much appreciated.

I’ve originally done all the tests you suggested jamesedge and yes, the rayhosts nodes are showing up correctly in the output window.
I will run some more tests to see if if it makes any difference on larger projects. I am wondering if a faster LAN network would make a difference? I am currently looking at a 10Gbe connection but I am not sure if it’s not better to to invest this cost in faster render slaves instead…

http://www.scan.co.uk/products/8-port-netgear-xs708e-100nes-prosafe-plus-switch-10-gigabit-ethernet-switch
http://www.scan.co.uk/products/intel-x540-t1-pci-e-10gbps-network-card-for-servers-and-workstations-cat-6a-copper

Thanks