aaronspink
Veteran
My point about fucking up in the past is that you can't really prepare for the situation of a Blizzard-level game launch in advance. Nobody can.
This is false. This will continue to be false. With the right back end design, you can scale upwards of 100 MILLION users on launch day without issue. Quite simply, it appears blizzard hasn't designed a reasonable modern backend.
There were ~4.5 MILLION copies in the wild on launch night according to recent figures and statements, and it's a reasonable guess that the vast majority of those copies tried to log in on launch day. That's an enormous load on the infrastructure.
THAT IS A JOKE of a load. Both EC3 and Azure could handle that without breaking a sweat, to name but two of many third party infrastructure providers. There quite simply is no technical reason why that load should ever be an issue.
You can do dry test-runs all day, and the stress test Blizzard executed reached 300k concurrent users (which is already significant; it's more than the full population of EVE online for example), but nothing short of a full launch can simulate a full launch, so how do you prepare FULLY for something like that? Can't be done, I'd say.
I'm pretty sure that given 100k$ and couple months, I could stress test it up to 10 million concurrent users. All you need is a simple network stimulator and hardware to run it on. This stuff is pretty basic in all honesty. Blizzard is just really really really bad about anything network and backend related.