Interesting post from WoW general forum.

[savaughn]

Nice numbers, but you didn't say what is causing the costs. Are you saying there's a lot of hardware failures, or are you saying you have to pay lots of techies lots of money to "support" the servers?

Edit: If you could point me in the right direction, I'd like to learn what makes this technology so inherently different from what I'm familiar with. Please don't simply say "high stress", I don't buy that by itself.
[right][snapback]71023[/snapback][/right]

I'd have to know what kind of systems you're familiar with to answer, but in general what we're talking about is streamed business logic requests from hundreds of thousands of simultaneous sources, virtually every single one of which needs to be parsed, turned into a back end database check, and the returned. This is going to be on the order of millions of events per second as your base load. This is then made substantially more complex by stream interaction. For example, if you have 40 people in visible range of each other, the stream complexity goes up by the square of the LOS, or 1600 times the front end server processing. Also, the database can be driven from multiple directions by simultaneous events (say, two people trying to mine the same node), meaning that you can richochet through business logic/dbase loops with event interrupts. (Loot lag seems so simple to fix when viewed client side but the server side part of that is down right frightening.)

I'd call that pretty high stress. By comparison systems like Google's character match index queries are snoozing.