[sldev] Re: [ARCH] Render unto Caesar...

[Argent Stonecutter]

On 18-Sep-2007, at 17:29, Mike Monkowski wrote:
> Argent Stonecutter wrote:
>> I really don't think that the architecture needs to be  
>> particularly  concerned with specific new types of assets, but  
>> rather that it  should be able to handle new types of assets  
>> efficiently in the  general case.

> But you need to know how much data is been transferred and how much  
> computation is needed in order to scale efficiently.

How much data is there in an as-yet-undesigned avatar mesh asset? You  
don't know.

"Premature optimization is the root of all evil." - Donald Knuth.

> The question is whether it is better to keep the avatar information  
> on specialized servers in the same domain as the region servers or  
> whether the region server should kepp all of the information.

The region server keeps as much information about the avatar as is  
needed to perform the operations on the avatar that the region server  
needs. That's true whether it's called a "region server" or a "sim".  
That's true regardless of the architecture.

You can't put information that the region server needs on a different  
server, because you will increase network traffic and STILL need to  
keep a copy of whatever it needs on the region server.

You can change the amount of information the region server caches,  
but that doesn't have a significant impact on the server  
architecture... it's all highly tunable.

> I'm trying to address the population limit on a sim.  I don't know  
> the best solution, but I know the current implementation doesn't  
> work.  Any ideas?

Multithreaded server using more than one core?