Interesting post from WoW general forum.

[Lissa]

Well I guess I was not clear. I didn't mean to dispute that the servers are high stress, I think it's pretty clear that they're under extremely high stress. I meant to say I don't see how that naturally requires high ongoing costs, after you have built and tweaked the system.

What I'm trying to say is that yes, it's a massive and highly complex system. But eventually, assuming no core system changes are needed, they should be able to get it running pretty smoothly with little interaction needed.
Let me try to explain what I meant by this.

And that's where the fallacy comes into play. You see, anytime you update code on a system, new bugs are introduced because either the bug was lying dormant and required some new code to come out or the new code clashes with some aspect of the running code and brings about the bug. Then we have other issues that crop up due to advances in the workstation hardware that can bring about still other bugs. In essence, the more you add to the system, the more complex it becomes and the more complex it becomes the easier it is to break somewhere which then requires manhours to go back and find what broke and fix and *hope* that your fix doesn't introduce another bug somewhere else.

I've dealt with servers that are nowhere near the same ballpark as the capacity or scale of MMO servers. But they're still computer technology, so they should have some basic similarities. I'd expect that their hardware does not simply die when put under loads that are too heavy, rather you'd get serious lag. I'd expect that their software may experience extreme bottlenecks too, which would also cause lag. I'd expect some crashes. But that's in the beginning. I'd also expect for the system to be able to be smoothened out eventually, to the point that its maintenance can be relatively automated assuming no core changes.
[right][snapback]71090[/snapback][/right]

That's another fallacy. The technology behind clustered servers is very much more complex and it isn't simply similar in anyway. The main reason for clustering is to increase uptime and add some fault tolerance, but even then, something can go awry (take the data center being hit by the tornado during beta). Due to the increased complexity, you have a lot more to worry about and need a lot more expertise in dealing with them than just being a normal server jockey. Calling it simply, the same technology, is not truly understanding the complexity of these beasts.

It gets even more complex in clustering depending on how your cluster is configured too as you could have several servers active at the same time while you have other servers that are passive waiting for one of the active servers to die. I was frankly amazed at how quickly they were able to bring the servers back up during the beta when they were hit by that tornado. Replacing servers in a cluster when one of them dies isn't easy and you hope and pray that while you are repairing the downed machine something doesn't happen to the other machines in the cluster (cause then you may be looking at rebuilding the cluster from scratch, a job that can take several working days). Clusters require a great deal of care to make sure that they continue to run properly.