PDA

View Full Version : Hypervoxels not rendering on some nodes



SeanCurtis
04-22-2005, 01:56 AM
The setup

Parent Server
-------------------------------------
Dual Opteron 244
2Gb DDR 400 ECC RAM
Windows 2000 Pro (SP4) installed on mirrored 36Gb Raptor HDDs (C: drive)
Lightwave installed on Raid 5 Array (5 x 300Gb drives, 1200Gb available) - (L: drive)
L drive is shared out with Full access to Everyone.

Folder Structure:

Root of L Drive
- Lightwave8 (program director)
- RenderContent (scenes/objects/images go in here)
- RenderOutput (frames are output here)

Command directory - L:\Lightwave8\Programs\Command
Configs directory - L:\Lightwave8\Configs
Content directory - L:\RenderContent

Child Nodes 1-10
-------------------------------------
Dual Opteron 244
2Gb DDR 400 ECC RAM
Windows 2000 Pro (SP4) installed on a single 36Gb Raptor HDD (C drive)
L drive mapped to the root of the L drive on the Parent Node (the nodes see the exact same L drive structure as the Parent Node)


The Problem

When rendering frames (we use TequilaScream) not all nodes render hypervoxels. I have read other threads where people say "perhaps exporting a hypervoxels particle motion .pfx file might help". I doubt this has anything to do with it or we'd surely see ALL nodes not rendering the hypervoxels. The only other things I can think may be causing this to happen is file locks for machines to access the plugins or something has somehow gone different on the windows install for some of the nodes (just stock win2k install with drivers, tequila node, mapped drive and vnc).

Any help would be much appreciated.

Phil
04-25-2005, 04:31 AM
I have seen this happen in LWSN mode III (local rendering, no controller) reproducibly. The same scene will render on the same machine without these problems in Layout, but pass it to LWSN and the HVs just completely fail to render, even in raytraced surfaces. I've mailed NT Tech support with content to demonstrate the problem, but never heard back. The HVs are the only things that go AWOL - everything else renders just fine so there is no config issue that I can identify.

Although you are running LWSN in a different mode, I wonder if the two issues are related. *shrug*

JML
04-25-2005, 07:32 AM
I had the same problem as you.
my hypervoxel would not render on my nodes with the new version of LW8

I later found out that it was actually the particle emmitter that was not emmitting anything..

to fix it, I saved the motion of the particle emmitter and it rendered just fine after that..

SeanCurtis
04-25-2005, 05:00 PM
Thanks for the replies guys.

Not sure if what you've said applies to our situation. We can render the scene and SOME frames have the HVs, some don't. The frames that don't have HVs renderred have everything else renderred ok, and it's always the same machines that don't render the HVs.

I'm checking now to see if it's a file lock issue on the plugin (windows 2000 only allows 10 connections at a time, we only have 10 machines so I don't know why we'd go above this)

Phil
04-25-2005, 05:59 PM
I should add that it is only sprite HVs that fail to render. Surface and volume HVs are rendering fine, but switch into sprite mode and LWSN fails to render them at all. Layout continues to render stuff perfectly. The scenes can be as simple as a null with HV applied and no raytracing or massively complex setups with millions of polygons and raytracing.....it's entirely reproducible.

It's incredibly frustrating.....

Hoopti
04-25-2005, 08:52 PM
Save your HV's out and they'll render fine, both on a network as well as a local machine. Just make sure that whatever directory you put it in is accessible via the network (ie the network drive is mapped and the same across all the machines, such as drive X:). That fixed my problem instantly. They render fine now.

Hoop

Phil
04-25-2005, 10:21 PM
Save your HV's out and they'll render fine, both on a network as well as a local machine. Just make sure that whatever directory you put it in is accessible via the network (ie the network drive is mapped and the same across all the machines, such as drive X:). That fixed my problem instantly. They render fine now.

Hoop

If you mean bake them, I've had very mixed results with that in the past. Baked HVs have never looked like the original ones in my experience and there's nothing in the documentation to explain why this is the case AFAIK. Another one of those in-your-face LW bugs that QA somehow manages to miss release after release *sigh*

SeanCurtis
04-25-2005, 10:27 PM
ok - I think some people are getting off the track.

The screamernet nodes render the hypervoxels ... but only SOME nodes render the HVs in the scene, not all nodes. The "some" are the same all the time (nodes 2, 3, 4 and 5). I'm wondering what's causing the same 6 NOT to render the HVs - file lock issues, plugin screamernet limits or some kind of difference in software between nodes (windows, updates, patches, drivers, etc).

I will post more information after further testing is done.

kevman3d
05-06-2005, 11:06 PM
The same machines... Hmmm... That's always a nuisance - Sometimes we've found a similiar issue on our nodes where some machines have the same problems consistantly.

A few things to try:
Check for errors - I would add the log output on the LWSN call (I think its -l from memory, but call LWSN on its own and it'll tell you pretty quickly) and review the logs to see if any errors occured during processing... That will tell you easily if there was a failure to load plugins, or read files.
Determine what's different with these machines (if at all) and possibly test one of them for issues with anything (like HDD, network, windows, etc). It doesn't hurt to see if perhaps you have three machines made with hardware from a bad batch of hard drives or other hardware - It does happen...
Make sure you have enough CAL's on your server to prevent network connection refusal on some nodes. With a 10 connection limit in general Windows, you run the risk of losing connection if for instance, someone on a workstation decided to connect and look around on the server for files - And check to make sure those nodes don't lose their connectivity at all for some reason.
We built a Linux file server with Samba just to avoid the whole network server connection limit and expense of buying CAL's - It works a treat, and a lot cheaper then Windows server and CAL's!
Just a note - If working with PFX you should *ALWAYS* save out your particle motion PFX files. There's no doubt about that fact - If you don't the PFX will be calculated on a per-node basis and could be randomised per frame, leaving you with jittering and odd looking pfx effects.
Run your scenes through LCSM and see if there were any issues with the scene that may be upsetting some nodes. (You can download it at the bottom of this page : http://www.newtek-europe.com/uk/community/lightwave/phillips/3.html)
I dunno if Tequila allows this, however in other controllers such as LightNet, you can optimise the farm to avoid sending jobs to nodes, and then reallocate the offenders to render other jobs that didn't have HV's.
Otherwise just remove the offending nodes from your renderfarm for this job - Avoid the problem to get the job rendered on time.

Good luck!