PDA

View Full Version : Network rendering so slow!



Thomas M.
08-25-2009, 01:00 AM
What might be the reason? Rendered in LW directly it takes about 5h, rendered via BNR4 it takes almost twice the time. Same computer, both 64bit with LW64bit, all 9.6. I'm just running to town to buy two i7, but it's depressing to get not all the bucks out of them.

Is Windows Vista somehow limiting the QuadCores to 50%?

I'm pretty angry about all the time I loose.

Anyone discovered something similar?

Cheers
Thomas

jameswillmott
08-25-2009, 01:10 AM
Are you using a radiosity cache that isn't being found properly by the nodes, and they are rebuilding the radiosity solution each time?

Thomas M.
08-25-2009, 01:45 AM
No radiosity at all. The times are about 6h compared to 8h. This might look not so bad, but in the end I loose hundreds of hours and that is too bad.

I just wonder why LWSN does such a bad job? Can it be related to BNR4? All cores run at 100%-95%, so the CPU is running at full speed. I don't know how LWSN could loose all this time otherwise?

down_in_black
08-25-2009, 03:48 AM
Have you tried to clear out all config and program files for net render? BNR usually keeps a copy of all these for all the net render clients.
When you render direct (locally) in LW you are using a different setup of configs and program files.

Could be worth a try?

Thomas M.
08-25-2009, 04:34 AM
I did in the past, but it didn't change anything. I wonder whether it is related to the settings from the computer with the LW folder as it copies all the settings from this machine. My BNR4 controller is not a LW computer, just a core2duo for taking care of the farm and all its cheep.

down_in_black
08-25-2009, 04:52 AM
Well I think it would be worth a try to copy files from your LW workstation just to be sure it is a good setup.

And if that does not work maybe send a question to BNR support.

Also check that you have the latest version of BNR.

Thomas M.
08-25-2009, 05:21 AM
Already contacted Paul and just wait for an answer.

JonW
08-25-2009, 05:58 PM
I found Screamernet sometimes takes longer than manually rendering from each computer separately, but obviously this is more work, but if there is weeks of rendering it may be worthwhile.

Since you are getting some new i7s, the 920 will give you the best frame output for the dollars per box. (3 x 920 are much better that 2 of the more expensive CPUs). If you haven’t already got an i7 you will be in for a pleasant surprise.

khan973
08-25-2009, 10:07 PM
Hei, I've hit many obstacles doing Network rendering and most of the time the problems come from us.

If you have more machines to do the job and it does it slower, the problem can be how the scene is managed.

1. Data access:
If the computer takes time to access the datas, it might take up to many minutes to get everything. So here are some things to be aware of:
- Gigabit network!!!
- NAS fast hard drive (or properly RAID configured one) so it allows many connections at the same toime without chocking
- Be sure all the datas are accessible
- The frames will be writen too, so the disk really have to be able to handle multiple access fast.

2. Textures
Try to manage your textures smartly.
- If you have video textures, use image sequences so it loads 1 image per frame and not the whole video each time. I faced that wit a 2 GO video to simulate water foam on an ocean.
- Never use bigger images than needed.
- PNG seem to be the format that's handled better by lightwave as it doesn't convert your 8 bits images to 24 bits image when loaded while it's the case with most formats

3.Config
You absolutely need a good configuration.
For example, If you have a bi-quad and 4 GO of Ram, it's gonna be a problem. As long as the node will load the scene, you'll have 4/8 GO which is 512 MO!!!
Imagine yourself working with 512 MO of Ram on your computer!
So you need at least 1GO per node. So using only 4 nodes on 8 can be faster in that particular case.
- Keep your Processors cool!


And last but not the least:
IT ALWAYS DEPEND ON YOUR SCENE!!!
The best configuration or node management depends ond your scene, how much computation needed, how many polys, what texture you use and so on...

But I hope my answers will help you.
I'm moving to a FLEX (Nec) solution with 6 racks (48 nodes) and memory management, it will help a lot on my projects!

Hieron
08-26-2009, 05:49 AM
Never noticed a strange slowdown due to BNR4 before..
You could do a manual LWSN setup and render to see where the problem lies.

Khan973: I'd love to build a blade setup once, but can't justify the added costs yet. I got the space to just put some midi towers with some cost effective consumer range CPU's (Q6600 before, i7 920's now)..

I have the feeling that a blade setup with the same processing power would sky rocket in price.. with a chassis and all..

Cageman
08-26-2009, 09:09 AM
We use Muster here at work and I also have it at home. The only time I've seen screamernet being slower is when using GI-cache. Don't know how BNR handles confings and such things, but with Muster it is quite easy to make sure the correct configs are loaded. Most of the time though, Screamernet has been slightly faster than F9 renders.