This is actually an OS problem and is the same as if you
had your page/swap file on the network. If you're using Linux with NFS and
caching enabled, it works quite ok since the network copy is transparently
copied to the local drive and mapped from there. The same cannot be said for
many other network drive implementations and especially not CIFS on
Windows.
- ½
With -p and -r the images will cache
even if they are located on a network drive. The problem is that MR will
constantly stream data from the map file as it's required for
rendering.
If you have a small subject within screen space that only
occupies a few buckets then this will not be an issue. But if the opposite is
true and you are rendering on a farm with lots of machines you will notice high
bandwidth demand on both the the file server and network.
With 100 nodes rendering full frame subjects with 5GB
of pyramidal / tiled textures the network becomes nearly
unusable and render times skyrocket. I've seen render times drop from 1+hours to
+/-10 minutes after re-locating all the textures locally onto the rendernode.
Votch-
