DBR seems inefficient

[Tommy L]

I am rendering a file using satellite (Distributed Bucket Rendering) and the efficiency is not what I expected. The render on my workstation (i7 hyperthreaded, so eight buckets) takes 14 minutes, with 1m43s translation time.

When I turn on DR I dont see the extra buckets in the frame, but the render time drops to 7mins (same translation time).

I am using DR over 7xQ660's, plus obviously the local i7.

Is this normal? THis is my first foray into DR with Mental Ray, Im a Vray guy. Any help much appreciated as always,

Tom.

[Matt McDonald]

You've got too many machines to DR a single frame. Mental Ray only allows six machines to contribute to a DR solution (counting the master - which should be the fastest machine of the lot).

 

Also, make sure that "use placeholder objects" is checked. This will help out the translation time quite a bit.

[Macer]

I've never tried DBR myself, as I don't have access to any computers that are really worth the effort! I know there is a limit with the number of CPUs used (8 sockets), but I don't think that is your issue. From the manual: (You can add up to eight physical CPUs. So you could add 8 single core machines, 4 dual-CPU machines, 8 quad core machines, etc.)

 

There is something about using IP addresses as well as computer names in this post http://forums.cgarchitect.com/38314-distributed-bucket-rendering-mental-ray-standalone.html

But I think that is aimed more at dual socket machines...

 

It is odd that the additional buckets don't show up. Can you check to see which machines are contributing?

 

Sorry I haven't been of any help at all! Perhaps someone with more DBR experience can help out.

[Justin Hunt]

DBR can be benificial but it can also be a pain in the neck. If your scene is really complex the time it takes to send the info to the other machines can negate the time saved using them. Having said that when it does work well it works very well.

 

I have been testing rending off HD frames using DBR, and it dropped the render times from 2 mins down to 45 seconds. unfortunatly its too unstable to use. I have been having issues with one slave dieing which causes a crash. I wish it could be that if a slave dies, mental just says, "OK your not working so I'll carry on with the other machines". I think its due to the fact the other machines dont have enough RAM and machines not clearing memory fast enough.

 

jhv

[datax]

First, if you got a lot of maps in your scene, make sure you check "Distributed Maps" and setup your satellites to access the maps directly. If you don't check "Distributed Maps" your master will send all the maps in the scene to every single satellite and I guess its using MR's native RAW format, so its uncompressed. So this might easily add up to several GB of data.

 

Beyond that it depends a lot on your scene. Simple scenes seem to start really fast on the satellites. Complex scenes take a lot of time to distribute to all the satellite, so they don't start to contribute for a while.

 

Do you have the "Show/Log Progress Messages" on?