deadline render manager and XSI

[Andi Farhall]

Anyone using dealiine on their render farm? We use it here and have been having problem where random nodes will never finish the frame that they are rendering - it happens on any scene - any version of xsi from 5.11 to 7.01 and resubmitting the frame to a different node and it renders. Our engineer that covers the farm just says "its XSI" but we never had this problem with batchserve which - I believe - does batch renders like deadline.

 

This means I have to babysit all the renders and resubmit frames - which is a right pita.

 

Any deadline users getting better behaviour?

 

cheers,

 

A.

[Alan Fregtman]

Out of curiosity, have you remoted into the machines when a frame is
misbehaving? Are there any C++ error dialog boxes?

There is an ocassional and extremely hard to repro nasty rendering
glitch which causes XSI to give a Visual C++ error dialog box, and
until somebody presses the OK button, the xsibatch process just stays
there, doing nothing. Not sure if Frantic prepared Deadline for that
situation.

[Andi Farhall]

Thanks for the info Alan, i'll get the engineer to take a look,

A.

[Matt]

Hey Andi, not familiar with deadline, but I have had this happen sometimes in Royal Render. The low-tech way around it was to have a cutoff point where if a frame takes longer than x minutes, it shuts it off and tries again on another machine.

Alan, have you managed to get rid of this C++ error? I have ran into this a few times and am curious to know how you have delt with it. It has been happening with scenes that are very memory intensive.

Thanks guys,

Matt

[Alan Fregtman]

Never managed to get rid of it, but a while back I made a simple
utility in AutoHotkey to wait in the background and close those error
dialogs if one comes up. At least then the process can die as it
should.

Compiled script and original source included:
http://s3.darkvertex.com/tools/AF_RuntimeErrorClicker_v1.zip

Hope it helps,

-- Alan

[Andi Farhall]

we have a similar cutoff setup but it resubmits the job to the same node - which 9 times out of 10 fail o render again. Out techy guy assures me were not suffering with the c++ thing - but then he also suggested that 2011 would fix it.