satchiel
11-02-2003, 12:05 PM
Hi, all,
I have been tasked with rendering a scene with Mental Ray on a small renderfarm of 5 computers, all x86, mostly AMD Athlon XP with various memory configurations, ranging from 256MB to 1 GB. Computers are connected via fast ethernet, 100Mbps, through a switch (I believe it to be a high quality one, if it makes a difference).
Maya 5.0.1 is used as a master and Mental Ray 3.2.6 is used on the rendering slaves. I ran a few tests and found that job distribution is very uneven. Most of the work is done by the Maya integration of Mental Ray on the master machine, with tiles from other computers coming in quite rarely, eventhough some of the rendering slaves easily outperform the master machine. More so for GI or final gather calculations. This is making the entire renderfarm rather inefficient (and at times even pointless).
Can anyone provide some insight as to why such a behavior is occuring? If we hypothetically assume that all computers have identical configurations, how likely is it that this is purely a network issue? Things like roundtrip time, ethernet collisions, etc ...?
I haven't yet tried variations using texture-intensive scenes as opposed to more heavily lit (final gather, GI) scenes. If anyone has moderate experience with network rendering with Mental Ray, I wouldn't mind any tips and tricks.
The only workaround I can think of is making tiles that the frame is broken into a bit larger, but I'm not sure if that is a user-definable option or if it's at all a viable solution.
Thanks in advance for any replies.
I have been tasked with rendering a scene with Mental Ray on a small renderfarm of 5 computers, all x86, mostly AMD Athlon XP with various memory configurations, ranging from 256MB to 1 GB. Computers are connected via fast ethernet, 100Mbps, through a switch (I believe it to be a high quality one, if it makes a difference).
Maya 5.0.1 is used as a master and Mental Ray 3.2.6 is used on the rendering slaves. I ran a few tests and found that job distribution is very uneven. Most of the work is done by the Maya integration of Mental Ray on the master machine, with tiles from other computers coming in quite rarely, eventhough some of the rendering slaves easily outperform the master machine. More so for GI or final gather calculations. This is making the entire renderfarm rather inefficient (and at times even pointless).
Can anyone provide some insight as to why such a behavior is occuring? If we hypothetically assume that all computers have identical configurations, how likely is it that this is purely a network issue? Things like roundtrip time, ethernet collisions, etc ...?
I haven't yet tried variations using texture-intensive scenes as opposed to more heavily lit (final gather, GI) scenes. If anyone has moderate experience with network rendering with Mental Ray, I wouldn't mind any tips and tricks.
The only workaround I can think of is making tiles that the frame is broken into a bit larger, but I'm not sure if that is a user-definable option or if it's at all a viable solution.
Thanks in advance for any replies.