Windows7 - render farm

[Soth]

Hi,
What modifications I should make to new Windows 7 installation on render slave (computer will be used only for rendering and will be on separate network (not connected to internet)).

Disable updates, windows firewall, disable UAC, set up fixed swap file (16GB?, computer will have 8GB of RAM), disable Luna, desktop image, set up optimise windows visual settings for speed...

What else, any ideas? :help:

Does LWSN still have problems reading from network mapped drives instead of network paths?

[rapscallion]

Hello!

I've just recently set-up my own render farm using an XP64 machine as the file server and application host, with two W7 machines as render nodes...

Believe it or not, this was surprisingly easy to do. The thing that really makes it simple is that you're using a closed local network with no connection to the internet - which means you can just disable all firewalls and allow discovery for all network devices/drives.

Make sure you are using a hard-wired Gigabit Switch between the server and the render nodes, as a standard 10/100 switch will be much too slow.

If your server machine has more than one ethernet port, you can leave one attached to the internet, but make sure that you have FIREWALL ENABLED and FILE SHARING DISABLED on the port. The second port connected to the switch can be set with NO FIREWALL and FULL ACCESS FILE SHARING.

As for Networked Drives or Network paths, it really depends.

When using BNR Render Controller, I found that a combination of both was needed; basically, map the network drives on the render node machines, but use full network paths inside BNR.

When using Lightwave locally on the render machines with control from the server through REMOTE DESKTOP, I found that mapping the network drives was sufficient - though certain plugins did not like being read across the network; to solve this, I did three things:

1. copied entire plugins directory to local position on each render machine
2. made unique configs folders specific to the render machines so that the locally hosted plug-in folders could be set permanently using a modified shortcut with the "-c" command; if you aren't familiar with making customized shortcuts with "-c", you'll really want to look into this, 'cuz it is very important and helps avoid many possible Lightwave issues when running the application from a single hosted location on multiple machines
3. launch Lightwave locally (from hosted position on mapped network drive) using the custom shortcut with unique designated configs folder, then scan the locally hosted plug-ins folder and exit the program to save the configs

One more tip: DISABLE WINDOWS ERROR REPORTING on the W7 render boxes...there are plenty of tutorials online for doing this, but here's a link to make your life easier:

http://www.addictivetips.com/windows-tips/how-to-disable-error-reporting-in-windows-7/

I don't know if any of that is helpful, but I can say that at this point, I have a fully functional BNR set-up, as well as two node machines able to run local Lightwave installations that I can control and submit scenes to through REMOTE DESKTOP.

Good luck!

[Adrian@Stufish]

Turn OFF windows bloody update !
After the buggers put out the 'chose a browser now' crap the other week, it jammed up the screens & prevented remote connections! I had to hare in to the office at sparrow-fart and replug everything into a kvm & manually click the 'F**k off' button on each of 20 client machines & re-start them.
Not Happy.

PS every machine in my office uses a common logon script to map drives from the server.
Ever machine with LW (or/and butterfly client) installed addresses the same config files on the server (just takes a bit of disciplin to remember not to change the configs without telling everyone, and remember to check the state of 'parent in place' and 'auto key' when you start a session!).
So everybody's scenes load the same stuff, plugins & presets and all, and the renders off the nodes look the same as the tests off the workstations!

[dnch]

i am thinking about similar setup..
one file server, couple of render nodes with cpu, ram and smaller disks.. (one or two for starters).. would ButterflyNetRender be any help?
i am mostly rendering large print files, lot of render passes (separated scene and object files).. i dont want to copy any project files to nodes (only plugins)

well.. sory for little offtopic, but i didnt want to start new thread

[Lightwolf]

T
Ever machine with LW (or/and butterfly client) installed addresses the same config files on the server (just takes a bit of disciplin to remember not to change the configs without telling everyone, and remember to check the state of 'parent in place' and 'auto key' when you start a session!).
You could also separate the plugin configs from the others configs. We had a shared plugin config and a server based lw#.cfg which was stored in a directory based on the current user name (so you can switch workstations).

Cheers,
Mike

[Roadwarrior]

Make sure you are using a hard-wired Gigabit Switch between the server and the render nodes, as a standard 10/100 switch will be much too slow.

What types of hard wired giga switch are you using or recommend?

[Roadwarrior]

Sorry i see its a Gigabit switch any others you recommend

[JonW]

One of my boxes, the onboardnetwork card stopped working. So I just put in a new card. I found this to be quick than some of the other build in cards. So I got a couple more. Some were quicker than built in and some were slower.

It’s trial and error. If you end up not using one card having a spare won’t do any harm.

The main benefit of /1000 switch, is when the scene is being loaded to a lot of nodes, especially if it’s a very large scene.

[djlithium]

Sorry i see its a Gigabit switch any others you recommend

I know it might seem costly but a good decent switch from dlink or even net gear isn't so much and you can burn them out on the cheap end of things.

I would go with the best un-managed switch you can get for the best price possible in the 8 to 16 port range (16 if you can, depending on your future needs of course and if you do blow a port, unlike in most hubs, the port only fails and not the entire unit along with it - so you have some hot spares so to speak). These can be anywhere from 80 to 350 bucks.

[djlithium]

Does LWSN still have problems reading from network mapped drives instead of network paths?

LWSN doesn't have any issues with reading from mapped drives really unless you are disconnected from the drive share or the server is rebooted and is not logged into again. Most of the time though you can reboot and the shares will automatically re-establish so long as there is no attempt to access them while the machine is unavailable.
I prefer to use mapped drives like x:\ or H:\ for my set ups rather than full UNC paths like \\LithiumSix\X\ etc etc.
Keeps file reads of plug-in configs and scene files nice and clean and fast and if you work like I do on a machine and your content directory is x:\katsmoviefxproject, the scene file is going to have that path in it for objects and images and all the rest. When you mix path types all kinds of things could explode and in general it slows it down anyway in my experience to do UNCs for configs and scene file reads.
But then again I am finicky when it comes to that.
Some people may disagree with me here, but make your life easier on yourself use a mapped drive.

[geo_n]

I had some trouble with win7 and using mixed os nodes.
"not enough server storage" problem on the nodes mapped drive with winxp.
This only occured in win7 as the server but a fix was to edit the registry.
http://winhlp.com/node/40

btw, i tried rendering on a 150mbps wifi. Not bad. No more wire clutter.

[Hieron]

"not enough server storage" problem on the nodes mapped drive with winxp.
This only occured in win7 as the server but a fix was to edit the registry.

Had those issues as well and it absolutely drove me mad. Getting that at 4am when masses of renders need to be done is NOT fun. I tried every " solution", just like that link and it never really solved it for real. (for my situation)

In the end my friend stepped in and changed the entire infrastructure of our renderfarm and set up an great Linux based fileserver, split VLANs over a big pro 24 port switch, bonded ethernet and alot of things like VPN server.

Since then, the shares worked flawless regardless of the amount of shares or nodes. I can log in from everywhere into our own network and work as if being local. So if I take the fastest workstation home, it still renders fine in the farm it's just not local (IT friends, priceless.... :))

[G@lvo]

Very useful thread!
Do you people prefer to utilize default LWSN or is better to rely on sw like BNR?
What advice would you give to one like me that is planning to build a renderfarm (regarding hw/sw-budget 8000 euros)?

[3DGFXStudios]

If you have just 2 nodes you don't need a fast switch. Even a cheap 100mbit switch doesn't slow your renders down. (just a little bit)

[Hieron]

Switches are cheap (50 euro for 8 port or so), why go for 100 Mbit? It will slow down loading times and if you use big GI caches it will slow down loading those alot too (done at renderstage). Copying alot of data around to a fileserver etc will be really slow too.


For 8k euros I'd get a nice 8 port Gbit switch, BNR (native screamernet = argh), and put together about 7 cheap but overclocked custom made i7 boxes. However, if you are not very RAM limited the new (early 2011) mainstream Sandy Bridge processors could be great as wel.
(ow and if possible get a linux fileserver to handle all the data and the mapped network shares. A linux box is supercheap, just need someone who knows what he is doing)

[G@lvo]

Thanks for the suggestions Hieron!
I'll talk to my boss and share my experience :thumbsup: