[fuso]

Hi guys,

I've been fighting quite a battle trying to use the cooperative render option after having updated to the latest version.
The behaviour is pretty unstable at least in that particular part of network rendering. Most of the jobs I have sent from
3Ds Max Design 2011 (with the latest plug-in) have failed for numerous and pretty random reasons and have caused me
a lot of trouble and grief at work.

Although I have been with maxwell from early alpha stages and I would consider myself as relatively experienced I don't
want to exclude the possibility of having made mistakes. For example yesterday I have sent 13 jobs in a queue using a
render farm of 9 PC's (Windows7 with identical specs). The first job seems to have finished ok, so I assume there is no
license issue as they all started off well. Then, when moving onto the next job the trouble started and numerous nodes
dropped out randomly, resulting the jobs to fail (no option of recovery as there were no temp mxi's to merge manually).
In some cases I could stop, remove and add the node back onto the job, but that hasn't always worked.

We currently have 9 licenses carefully installed where we need nodes or full standard versions which has always worked
just fine. Our IT group assured me that all nodes have full network access and that our network is running at reasonable
speeds for using this sort of service. Again, all drives are mapped so there's no need to send dependencies, I guess.
All the jobs were sent using the same settings: cooperative mode, low CPU priority, no dependencies, small resolution and
low SL. In the next step I tried both options: use any node, and use slected nodes only with the same result, failures.
I have copied a few lines form the manager/node logs as follows:

[18/January/2011 11:09:28] Error binding render node to port: 45457
[18/January/2011 11:09:28] Another instance of Maxwell could be running
[18/January/2011 11:09:28] Tcp server can not listen port: 45460
[18/January/2011 11:09:28] Listening in : \\.\pipe\maxwell_network_connection
[18/January/2011 11:09:28] Searching manager...
[18/January/2011 11:09:28] Broadcasting using port: 45456
[18/January/2011 11:09:28] Maxwell node 2.5.0.0 successfully initialized.
[18/January/2011 11:09:28] Temp folder created in: C:\Users\VISUAL~1\AppData\Local\Temp\mxnetwork\rendernode\
...and
[18/January/2011 11:29:10] Rendering process has failed in node: JW222: 192.168.1.204
[18/January/2011 11:29:10] Render node JW222 has reached the maximum number of failures permitted and it will be closed.

Furthermore I have copied details from the Windows7 crash report following the last log entry of one crashed node:

Problem signature:
Problem Event Name: APPCRASH
Application Name: mxnetwork.exe
Application Version: 0.0.0.0
Application Timestamp: 4cf8f936
Fault Module Name: QtNetwork4.dll
Fault Module Version: 4.5.1.0
Fault Module Timestamp: 49ff14ec
Exception Code: c0000005
Exception Offset: 0000000000067713
OS Version: 6.1.7600.2.0.0.256.4
Locale ID: 2057
Additional Information 1: 68c0
Additional Information 2: 68c0990361c7f99bd5313f7563199610
Additional Information 3: 4106
Additional Information 4: 4106f2dee0b7e770f00458927ec5150f


Please let me know if there's anything I've done wrong or if you have a clue what's going on. As I've never had a version
this unstable I hope someone can advise and hopefully the problem will be fixed soon.

Thanks a lot in advance.

Jost

[Bubbaloo]

I haven't net rendered on 2.5 yet, but I have found in the past that the most consistent way I get good results is to use UNC paths instead of mapped paths, and always send dependencies.

[fuso]

Thanks for the info Brian,

I'll do a test with the same files but checking 'send dependencies'. I'm just worried what happens to all the files copied
across after the render is finished. As the PC's in our render farm are actually workstations of other employees I don't
want to clog up their hardrives unnecessarlily. Will they be deleted when you reset the network/selected nodes or when
you kill the nodes?

Thanks,

J

[Bubbaloo]

As far as I know, the files are stored in a temp folder which is deleted after the render.

[fuso]

I haven't net rendered on 2.5 yet, but I have found in the past that the most consistent way I get good results is to use UNC paths instead of mapped paths, and always send dependencies.

Ok, done that test and nothing has changed. In fact, it seems to get worse somehow. Now I also get error messages
about the maximum number of licenses reached which is nonsense as we have 9 installed on 9 PC's. The cooperative
network used to be such a powerful and reliable tool in V 2.1 so the only things have changed is the version itself as well
as all our PC's have been switched to Windows7. Am I really the only one out there having this kind of problem? I could
really use some help here, otherwise I have to roll back to V2.1 which would be a nightmare.

Thanks

[Bubbaloo]

Make sure you contact NL through the customer portal. I'm sure they will get you fixed up.

[brodie_geers]

I don't have much experience doing network renders on 2.5 as of yet but I've had some issues with the network render for awhile now. They aren't quite the same as yours but the one similarity I see is that you said the first rendering works fine and then after that things go to pot.

An odd workaround I found that pretty much fixed my issue was to not use the Stop button in the monitor to end a rendering. Instead, make sure that the rendering stops on it's own by reaching the target SL. That means either just waiting for it to finish, or if you want to stop it early you can edit the target SL to be something less than what the current rendered SL is at.

It's not much, but it's worth a shot perhaps.

-brodie

[frosty_ramen]

One issue that i have had is that multiple users were signed into a computer that was being used as a node, when the second user would log in (either physically or remotely) the node would start. This happened because i had the node in the startup.

[micheloupatrick]

I had the "maximum number of licenses reached" error with 2.5, and I solved it (or so it seems) by leaving only ONE network interface active on each machine : that means turning off the Wi-fi interface, the second ethernet port on a Mac Pro, etc...
But as far as I remember, I never had this error before the 2.5 version.

[dariolanza]

Hello Jost,

I would suggest you to submit an scene that doesn't present any texture (the Cornell.mxs test scene may work fine for this). This will help us to isolate if the problem is regarding the texture paths.

If you can reproduce the fail, enable the "Deep Debug" in the Manager, Monitor and RenderNodes preferences panel (on each console, File>Preferences and enable it on the Minimum Log Verbosity dropdown).
This will enable a higher verbosity level that will display more information about any problem next time it fails.

Let me know the evolution of this test.

Greetings

Dario Lanza

[fuso]

Thanks for your input Dario,

We discovered that we had our 3 standard license file installed on 4 PC's instead of using the node. I am not sure if that
could have been the problem as I thought that it wouldn't matter where what license file is installed if you only render via
network. However, after we fixed that, it seems to work ok, even with files using plenty of textures all over the place and
disabling 'send dependecies'. In fact, it seems the whole network is a lot more stable now. We also decided to only use 8
out of our 9 licenses, in order to be more flexible. Not the ideal solution, but it works for now.

If the problem persists I will take the time to run the test you suggested, but due to deadlines I can't do that right now.
But to be quite honest, I don't think the current release deserves something like version 2.5. I am with Maxwell Render
since the beginning and I still feel a bit 'cheated on' as previous releases seemed to have been a lot more stable
altogether. Nevertheless I am looking forward to more stable releases in the future. For some reason my fascination for
this render engine is still pretty much unspoiled.

Thanks for having looked into this anyway, I appreciate that.

Jost