Local Test Renders are Quick - Network Renders are Much Slower

[MegaPixel]

Using 3DS Max 2009 64bit w/V-Ray 1.5 SP4

I've noticed that when doing final test renders on my workstation, that the render times are much better then when it is participating in the network render of the same project (In this case, a 3x performance hit). This is an animation project using the IR prepass and rendering method. Could this be caused by excessivly high interp. samples and interp. frames values for the IR map during the "rendering" phase? (see attached IR settings below).

How do these settings affect the "rendering" phase and should they be different from the values used during the "prepass" phase?

Thank you in advance for any help - Mega

[Dave_Wortley]

If all your render blades are sourcing the LC/IR map off the same network location then it will slow down your overall rendering, because of fighting for read/write speed.

[GLASS-CANVAS]

What's your network rendering setup? (number of machines, storage system etc)

[MegaPixel]

My Backburner Manager is currently on a 5 year old Dell running Windows XP 32 - Just a 5200rpm IDE hard drive. Network is all gigabit speed. The Farm PC's (roughly 8 at the moment) are mixed Win. Xp Pro 64 and Win. 7 Pro 64 and are much newer (from 2 years old to present day), but much faster then the Manager PC with drives speeds from 7200rpm, to 14000rpm.

So this might not necessarily be a V-Ray setup issue, but possibly a network setup or manager hardware issue? For this particular project, I was pointing all servers to the Manager PC for the project files (Source files, Precalc files, and Final Frame files). This method seems to have worked fine for me in the past yet somehow I think there was always a performance difference between my local test renders and network participation renders.

I'm not sure it would make sense to copy the source content and the precalc files locally to all servers would it? - I mean once the project is opened, the servers don't have to pull it again until the job is done, right?. The only thing the Servers should be doing is pulling a few IR precalc maps to start a new frame, then save that frame when it's done - rinse repeat. The frame times on my servers range from 5 minutes a frame to 20 minutes a frame so it's rare that all or even half are trying to hit the Manager at the same time.

How do you guys have your networks setup? I'm sure there were budget limitations which drove us to run things this way.

Thanks - Mega

[cheerioboy]

With my small three machine farm, I notice slowness at the beginning when all the machines are acquiring the precalc files. After that they're pretty steady although there is some interrupt from all the machines saving to a network drive. That doesn't affect the rendering time so much as I takes longer in between frames to get going. So sometimes if the sequence is a mask pass or something that renders quickly. It will render faster if I run it locally to the local drive, then transfer.

[MegaPixel]

So aside from hard-drive speed differences, does noone see any performance issues with my Interpolation Samples or Frames values for the job? (attached above) I'm just not ready to believe that my network hardware setup is so bad that I'm seeing a 4x performance decrease - Surely I'm screwing something up on my Job settings somewhere.