network rendering on Mac OS X 10.6?

[zeitmeister]

Hi,

I am doing a network rendering with one master and two clients at the moment, and I am noticing some strange things.
I am using the official Indigo 2.2.12 on all cients and master, master and one client running on Mac OS 10.6, one client on Mac OS 10.5.8.

At first, Indigo shows the connected clients properly. Both clients got the scene and are rendering.
BUT: One single "frame" has got 2.639 GB. The clients say that after rendering, the frame has been uploaded properly, but the time for that has been 0.00 seconds.
This is strange, because I didn't connect the Macs via fibre channel.
At my master GUI, in the network rendering tab, both clients appear properly, but it shows that they do 0 samples per second.

I don't use a proxy, if this could be one reason.

Is something going wrong there, and what and why?


Thank you very much in advance!

[zeitmeister]

Okay, I tested it with a smaller scene: frame size 90 MB.
Everything working fine.

So that means that something can't handle this large framesize. I suppose the following:
The master plus the clients are refreshing more often that it takes them to send the frames to the master.

Now my question:
Where can I specify the refresh/upload/transfer time?
If it takes 4 minutes to transfer the 2.93 GB to the from the clients to the master, I have to set the refresh/upload/transfer time to something explicitly higher.

How can I do that?


Thanks in advance!

[OnoSendai]

Hi Zeitmeister,
I think there may be an issue with transfer network frames larger than about 2GB. I'll try and fix this soon.

You can change the frame_upload_period to something larger in inifile.xml. The value is in seconds.

[zeitmeister]

Hey Ono,

did I place it correctly?

Code: Select all

<?xml version="1.0" encoding="utf-8"?>

<config>
	<renderer_settings>
		<metropolis>false</metropolis>
		<bidirectional>true</bidirectional>

		<downsize_filter>
			<mn_cubic>
				<ring>0.2</ring>
				<blur>0.6</blur>
			</mn_cubic>
		</downsize_filter>

		<super_sample_factor>2</super_sample_factor>

		<aperture_diffraction>false</aperture_diffraction>
		<post_process_diffraction>true</post_process_diffraction>
		
		<frame_upload_period>400</frame_upload_period>
		
	</renderer_settings>
</config>

Tested it now, and I still got the same message on the clients: Frame uploaded, transfer time 0.00 seconds.

Seems like it's really about the large filesize.
Did you crop that because of the FAT32 filesize issue?

[OnoSendai]

You got the right place for the frame upload period.
But the 2GB limit is a separate issue. For now you can't use frames > 2GB on the network sorry!

[zeitmeister]

Don't mind.

But please think about fixing that; Indigo network rendering is exactly supposed for these cases... rendering big resolutions for banner prints etc.

Do you think this is issue is easy to solve? I think that PNG is able to go greater that 2GB.

[Zom-B]

Do you think this is issue is easy to solve? I think that PNG is able to go greater that 2GB.

Its the IGI that gets transfered, not a 8bit image file

How about this: http://www.indigorenderer.com/forum/vie ... f=5&t=9961

[OnoSendai]

It's easy enough to fix, I'll have a go at it tomorrow. I fear for your network however

[OnoSendai]

ZomB: bucket rendering is indeed a nice fit for network rendering. I'll think some more about it.

[Zom-B]

I'll think some more about it.

Thats all I want, thanks a lot master Ono

[zeitmeister]

Ah, the IGI file is relevant, not the megapixels or whatever.
Ever missed that.

Ono, my network has to do a job, so no mercy!

I noticed that the 2 GB-frontier is not precise enough... on the Mac, perhaps.
I had to reduce the framesize to 1,861 GB (as displayed in the clients GUI); everything above (1,9 GB for example) didn't work either.
Is this a rounding issue resulting from the different file package sizes on HFS, NTFS or FAT32 systems?

And another question:
Does the client stop rendering while transferring the frame to the master?
So does it makes sense in general to force some kind of "respite" with the manual <upload_frame_period> method?

[Zom-B]

Does the client stop rendering while transferring the frame to the master?

Afaik the client keeps computing samples during transferring the igi over network (CPU usage spays at 100% for each core, and transferring a file should be this much of a CPU Killer).

As long as this ~2GB limit exists, you'll have to manage the igi size by wisely using SuperSampling & lightLayers :/

[zeitmeister]

Hm, in this case I haven't got any light layers except sun and sky, ans supersampling is 2.
So even less than ordinary in my projects.

[Zom-B]

So your resolution seems to be quite high ^^

Anyway, if not needed, you can merge the sky+sun Layer.
Slimly uncheck the option in Cindigo under "Indigo render Settings > Environment" to save 50% of IGI filesize

[zeitmeister]

You're kidding! For real?!?

Gonna check this immediately.