ElCadaver wrote:I was considering saving up to build one of these. Do you think it would make a good NWN2 server? Just an idea mind you, if I ever got the time to get around to it. Cheap as chips.
http://www.calvin.edu/~adams/research/microwulf/
Oh i wish nwn2 and other regular proprietary software could scale like that
Basically the bottleneck with NWN2 isn't hardware. Any single module cannot address more than 2gb memory,- simply because Obsidian does not offer a 64bit binary of the server.
The i386 architecture does not allow for addressing more than 2048mb ram to any single running process.
Thus the only way to get this to work is running multiple modules, and setting up portals between them. If you buy say a quad core machine with 8gb ram you could basically host 3 modules on it at the same time, and have say two of those modules contain large exteriors with lots of detail.
So basically the point here is to stop thinking about a module as a server. When we start to redefine what a module is, we can do all this stuff.
No reason to set it up as a beocluster or anything like that though. Nwnserver2 is single threaded and wouldn't benefit from the parallelism of a cluster,- simply linux on the bottom, and a vmware session for each of the 3 remaining cores, with 2gb ram each would be sufficient to host this. You would probably want a raid 0+1 though, so thats four drives to get the necessary IO throughput + backup.
You can get a machine like this for under the price you mentioned though,- as DDR2-667/800 ram is super cheap now and more than good enough for this purpose,- same is true for an old intel 6600 quadcore,- an of course you only need on board graphics just to install the underlying OS.
Just for info: The current BG module takes about 670mb ram when loaded,- the TSM one about 1,5gb. Since you want some overhead on top of that, TSM would likely need to split modules if it keeps growing,- especially with exteriors.
PS: Very cool machine though. Imagine doing it with 4 mobos with a quadcore on each and say 4 times the memory
In case you didn't know,- machines like that are good for workloads like: seti@home, encoding/decoding, compiling code, rendering 3d cinematics, cracking encryption, etc etc. Basically huge math workloads.
For scaling stuff like serverdaemons though, virtualization is king now...
