Often people ask me what are we using? Some of it may like the Colonels secret recipe but not all of it is... So what are the 'Region Servers' - those lurking beasts in the Datacentre that run simulators running the regions. Basically we have 3 types, in trying to find the optimum balance for the requirements of the code, the load & efficiency.
Let's start with the newest Servers being added to the grid at the moment.
Type 1. Dual Harpertown Xeon's (8 Processing Cores)
In most deployment scenarios if you have an application that loves to eat RAM then it will benefit greatly by increased processor cahce. These new bad boys are Intels latest to the Xeon family with a combined total L2 Cache of 24Mb.
So... 2X 5410 Xeon's 2.33Ghz a core 4 Cores 1333Mhz FSB - Total 8 Processing cores. using 800Mhz FBD's (RAM)
These servers are based on 45nm technology, and are cooler than their counterparts, using less power (but more processing output).
For the boffins here's 2 links that may interest you:
http://communities.intel.com/openport/blogs/server/tags/harpertown
http://communities.intel.com/openport/blogs/general/2007/09/17/harpertown-45nm-under-the-hood
OK What about No.2?
The second type of server we use are Dual Clovertown Xeons. These are 5320's 4 Cores again per cpu so 8 cores a server, 16Mb total cache 2x4Mb x2 Cpu's. They run at 1066Mhz FSB and a core speed of 1.86Ghz per core.
Don't be mistaken these are still blazing servers and less than a month old, and it's worth mentioning none of the servers we use struggle as an Openlife Simulator, actually there's plenty of headroom to grow. 
Server Type 3.
These are kewl -> Powerhouses of fun. Here the difference is we load them with less Simulators but they have more memory freedom.
Quad Core Kentsfield Xeon Servers.
Intel Xeon 3210 Quad Core's running 2.13Ghz, 1066FSB & 2x 4Mb Cache.
These server's have plenty of grunt and on a 'core' level perform slightly quicker than the clovertown core in performance testing. As we exceed the requirements of the simulator already this doesn't translate to anything visible in world but is another one of the factors to consider for a long term standardized simulator platform.
So does this automatically mean that server type 1, is the automatic winner? It seems the biggest... Not necessarily, it's important to not that 8 processsing cores in all it's glory doesn't exactly equal one 'Awesomely quick simulator', the code doesn't have great support for multi-cores yet, so a region will not make any significant use of 8 cores over 2 cores. And even 'cores' themselves have their own advantages over each other.
We'll continue testing with these in the grid, and as the code develops further and the load increases I'll post more information, perhaps next some pics? For those who like looking at shiny expensive metal & silicon?
Sakai Openlife 