|
Hi all,
First of all apologies for the long thread. However the fact that DBR fails for so many of us makes this a valid and, i feel, important thing to discuss and try to solve.
As others have reported here (and elsewhere) many (most? all?!) of us are having issues with MR crashing during DBR. Sometimes when you hit render everything works, sometimes MR crashes half way, sometimes it crashes at the “cleanup” stage. Whatever the case it’s extremely unstable and makes using DBR in production on a tight deadline (or any deadline for that matter) completely unpredictable and therefore unfeasible/unusable.
to be clear before we start, for me at least it is MR that is crashing - “fatal error”, “unhandled exception” etc, NOT MAX. However the MR error window prevents you from using MAX once it’s popped up, so you cannot manually jump into the framebuffer and save out the finished image. upon accepting that MR has died a nasty death the whole load closes down, perhaps with the option to save a backup version, perhaps not… again this is another inconsistency with the crash.
to be clearer, the crash for me when using DBR can occur either immediately after render launch, during a render at any point or after a render when “cleanup” is in progress. Interestingly the last situation (cleanup crash) is the most common for me. the crash using DBR occurs using GI, FG, both or none, using BSP2, the original BSP, any of the other BSP settings. All scenes render perfectly on the master machine.
my setup (for reference):
3 machines total including master server machine.
All three machines are the same barebones (Shuttle SX38P2Pro, same motherboard, PSU, same RAM brand etc).
2 of them are Q9450 CPU (quad core), 8GB RAM, 500GB HDD, GeForce8600GTS.
1 of them (the main server machine) is QX9650 (quad core), 8GB RAM, 1TB HDD, GeForce8600GTS.
(Basically all three are the same - 12 buckets in total).
Gigabit ethernet LAN using Netgear 5-port switch with 10m CAT5E cables.
MAX 2009 Design 64-Bit on master, nodes used only for DBR obviously so just raysat running on those.
Windows XP 64-bit on all 3 machines.
I propose that we all chip in with our tips on setting up network/MAX/MR for DBR in the hopes that there is a possible “best practice” to get this to work as often as possible (or even all the time!).
I will start:
- hardware - for the best compatibility try to use machines that are all the same or similar spec. the most important being that they are all Intel or all AMD (i think procedural textures and other things like this render differently across CPU types so it’s important to use all Intel or all AMD) - all three of my nodes are identical down to graphics card, RAM and CPU so this should not be a factor.
- windows version - make sure all your nodes are running the same version of windows, pretty obvious. all of my nodes are on XP64 so this should not be a factor.
- same version of max - pretty obvious! they should all be the same version number and all 32 or all 64 bit.
- memory - the obvious first step is to ensure your nodes have enough RAM to cope with whatever scene you are throwing at it. If your “main” machine has 8GB but your nodes have only 4GB but your scene requries, say, 6GB to render, your DBR is going to fail. I have 8GB on all three of my nodes all the same speed so RAM should not be a factor.
- network speed and quality - i have a Gigabit LAN with a brand new Netgear 5-port switch and CAT5E cables (short 10m cables). speed and network reliability should not be a factor.
- general network performance in windows - make sure you can actually read and write and generally talk to all your nodes within windows itself. make sure each machine can “see” and access (read AND write) to your “main” machine (the server). this is the case with my setup so should not be a factor.
now, as far as i’m concerned at this point everything should just work perfectly using DBR. all the nodes are the same hardware and software config, all on a fast, reliable network and all talking to each other just fine. if the scene renders on the server machine it should render exactly the same on the nodes, no?
there are a few things I’m considering that might cause the inconsistencies with DBR success:
- softimage XSI (with MR of course) used to (often) require you to set up a separate “render” user account in windows in order to let DBR work properly. perhaps doing something like this for DBR with MR under MAX might help the situation? I doubt it but it’s a possibility.
- this “cleanup” stage. not sure what this actually is or why when DBR fails it is often at this stage (the whole render completes perfectly, then the “cleanup” begins and boom, MR crashes). if it’s only MR that’s crashed here why is MAX not responding afterwards? Would it not be possible at this stage to somehow prohibit that message from blocking MAX so you could actually manually jump into the frame buffer and save out the finished image?
- network cable quality - for sure good quality cables are a must, but i would imagine CAT5E would be more than adequate for this?
- mr’s “processing” options - those three checkboxes and the memory value. Could somebody indicate if these would make a difference at all or are likely to be a factor here?
- maps and other “info” that needs to be sent to each node - surely DBR is purely a “CPU-based” network rendering solution, so maps and other scene items and their paths should not be an issue here?
I look forward to hearing your thoughts on this as it’s been an issue for me since MAX version 9 (didn’t even try it before that) making DBR ultimately useless for me and many others it seems.
Cheers,
alex york
for and on behalf of Atelier York | Bespoke Architectural Visualization
http://www.atelieryork.co.uk
MentalRayTips Twitter Feed
| 
