PDA

View Full Version : Mac Screamernet for Beginners



awrieger
04-23-2007, 01:47 PM
This is a long reply I made on another forum, but I thought I'd cross-post it here in its own thread (so with slight editing for context) as well because of the dearth of info I found about setting up Screamernet on the Mac while researching it last week. I just set up a LW Screamernet network of old Macs myself over the weekend, so everything is still fresh in my mind.

I suggest Renderfarm Commander for Mac OSX (http://www.brucerayne.com/renderfarm.html) to set up and control your render network. My eyes glazed over and my brain went numb when reading the PC-only tutorials and instructions on setting up LW's Screamernet network rendering. After two days I was still totally clueless, so I bought Renderfarm Commander for the Mac for AUD $57 and had an 8 machine network of really old Macs up and rendering within an hour.

It works with both Mac and PC networked machines, but it automatically creates the node data for Macs, whereas you still have to set up the PC nodes manually.

I bought the 'hobby' version which allows for 8 network nodes (CPUs) maximum. If you have a multi-CPU machine, run one node per CPU. You can get the free version which allows only 2 nodes and see if you like how it works before buying the hobby version. Then there's the 'pro' 500 node version for an extra AUD $60 upgrade, total AUD $120 all up.

If you want to run 500 nodes, you'll need to switch to Mac OSX Server (unlimited clients version) as the standard Mac OSX only allows for 10 machines to be connected at once (the same as the base Mac OSX Server too with only 10 clients).

Renderfarm Commander also does split image renders (renders larger images into smaller 'limited region' blocks of from 2 to 16, so the max number of network machines used will be 16), and automatically assembles them together into the one large image.

To give an idea of speed boost using even obsolete slow computers when doing network rendering, a G4 DP 1.25GHz renders a 60 frame animation @ 3 minutes/frame = 180 minutes = 3 hours. Add 6 even older networked G3 400MHz Macs ($30 each off ebay) which render the same frame three times slower @ 9 minutes, so they render out 6 x 1 frames for every 3 frames the G4 renders out.

So even though each G3 is 3x slower than the G4, six of them together actually give a 200% speed boost over the G4 alone, taking the total render time down from 3 hours to just 1 hour.

So sticking any old Mac (running OSX) you have sitting around or can pick up on ebay for a few dollars onto a network will help. The key is the number, the more you add, the faster the complete process. eg, 60x G3's will render all 60 frames in 9 minutes.

Of course everything will be MUCH faster with newer machines, ie a row of quad G5s or a stack of 2GHz Mac Mini Core 2 Duos (when they come out), but until Lightwave is released as a universal binary PowerPC Macs are the way to go as it performs extremely sluggishly in Rosetta emulation on Intel Macs. My Macbook 2GHz Core Duo rendered even slower than any of the obsolete G3 400MHz machines when I added it to the network. So for the same price and performance of the Macbook rendering 1 frame, you could have 100 G3s rendering 100 frames in the same time period (physical space is the only limitation really).

Anything else you need to know is set out in the easy to read step-by-step Renderfarm Commander instructions pdf file. Happy Rendering!

NB. FPrime doesn't work over network rendering yet unless it can launch a dongle-less copy of LW Layout on each networked machine, ie the demo version. And it's not universal binary yet either.

NB2. On non-altivec cpu's like G3s, set thread number per node to 1. On altivec cpu's like G4 (or G5), I found rendering was fastest for my specific test scene when setting threads to 4 per node. Perhaps a G5 can handle even more threads. Different types of scenes may also affect the efficiency of the number of threads per node.

PS. I recommend 'Chicken of the VNC' (freeware) to control all the other Macs remotely from the main desktop Mac. It's sort of a lite version of Apple's 'Remote Desktop'.

awrieger
04-23-2007, 02:10 PM
I should note that my speed comparison between the Dual G4 and the G3s is based on using only one of the G4's processors as a render node, leaving the other processor free to carry on with general computer work.

This effectively allows 'background' rendering while still working in LW.

(If I wanted to use both the G4's processors as render nodes on a network, I would simply start up 2 LWSN nodes on the G4 to run at the same time).

awrieger
04-23-2007, 04:13 PM
Update: I can't edit my original post, so please disregard the nonsense comparison I wrote about the extremely sluggish speed of my Macbook because of LW operating in Rosetta emulation when connected to the network.

It appears this was actually due to my simply using the wrong type of ethernet cable to connect it. Crossover cables are a no no, apparently - use one and you will indeed make a new Macbook as slow as an old G3 over a network hub.

bassmanjam
04-25-2007, 12:06 AM
Excellent info. I agree with the Mac Mini idea - decent performance with a small footprint.

Thanks for posting your findings :thumbsup:

v1u1ant
05-17-2007, 05:40 AM
yea thats really helpful thanks