distributed rendering questions

[posterus]

hello,
i just recently started messing with distributed rendering and have a few questions forgive me if they've been asked already...i'm not seeing the answer in the help files.
first of all, DR is slower than just using my single workstation. my test setup consists of my workstation (macpro) and 3 servers of mixed configs. see below. my test scene is using IR+QMC. one thing im noticing is during the IR calc i only see one bucket for my macpro howver shouldn't ther ebe 4 buckets since the machine has 4 cpus? When it begins to render the image i can see 4 buckets for the macpro. The same is true for the other multi-processor machines. when in DR mode does it condesnse multi-processor machines into a single bucket during IR calc??
also using DR ended up being 6 seconds slower than just using my single workstation?? any ideas why? i'm sure its something simple
thanks,
-joe

[dlparisi]

The missing buckets during DR are due to it simply overwriting the white box while it updates the irradiance map. It's still working on the bucket and updating it though (Vlado has stated this is unavoidable). This only occurs with the Max FB (which you're using), switch to the VRay FB and they'll show up correctly.

I had similar results when trying to include a slow machine in a DR calc (in my case a P4 3ghz) which couldn't hold it's own to a dual opteron 270 paired with two core2 duos. I deleted the P4 and the times went up. Seems the time spent waiting for this one machine to render a bucket is greater than just leaving it out of the calcs and letting the other machines handle it. I suspect your "Chicago" machine may be the culrpit.

[posterus]

thanks for the tips. i removed the slow server and it rendered in 56 secs. i enabled the vray FB so i can see all the buckets. it still seems inefficient in the IR calc. for the 1st half of the IR calc only the 4 buckets from the macpro show up. eventually the other servers show up and do a few buckets. is this normal? why dont all buckets from all servers show up as soon as IR calc begins?
thanks
-joe

edit:
continuing to test...this doesn't add up. im using our 3 fastest machines for DR, our macpro and 2 dual core boxx machines. With a new scene, the macpro alone beats the macpro+2 boxx's by 2 seconds. its definately not using all of the available cpu power in DR. do the workstations and the servers need to have similar specs in order to be efficient?

[vlado]

Due to some problem with 3dsmax 9 (it always tells V-Ray that multithreading is OFF), you will need to explicitly force V-Ray to use all available cores by opening the maxscript listener and typing:

renderers.current.system_numThreads=4

For later versions, there is no need to do this, as V-Ray uses automatically all cores regardless of the multithreaded option in 3dsmax.

Best regards,
Vlado

[gfa2]

why dont all buckets from all servers show up as soon as IR calc begins?

I find that scenes with a lot of geometry take a while to get copied to all the DR machines and that's why they don't start right away.

do the workstations and the servers need to have similar specs in order to be efficient?

I'd say no. I use everything from dual dual core Opterons to 2.4 GHz P4 on the same DR renderings and haven't had any problems. Is your network stable and fast?

[DaForce]

Depending on how slow your slowest machine is it may well be quicker to not have in working on the DR.

Say all the fast machines finish and the slow machine is slow crunching away at its last bucket. It would have probabaly been quicker to not have that machine there at all.
But obviously that depends on a few things and may only be relevant in extreme cases.

[harperperkins]

something that worked well for me because of the varying computer specs on our network was to lower the bucket size, like down to 16. The smaller buckets allowed for the slower computers to not have to crunch as many numbers to finish their bucket, and would allow the faster machines to speed through theirs.

ME

[TRGraphics]

something that worked well for me because of the varying computer specs on our network was to lower the bucket size, like down to 16. The smaller buckets allowed for the slower computers to not have to crunch as many numbers to finish their bucket, and would allow the faster machines to speed through theirs.

ME

I did the same thing and it made a sizable difference in render times. I was able to put 2 computers on the DR system network that I had taken off because they actually made the render slower.

[werticus]

so:
smaller buckets
vray frame buffer

then press f11
and paste:
renderers.current.system_numThreads=4
enter

and your sweet