Let's run the numbers. Let's figure a 50 ship fleet. Average 7 night cruises and the same criteria.
W'ere now at 125000 passengers/week, 6.5 million/year (obviously less - those 50 ships aren't running full, but let's go with it). 3.25 million passenger vists. 48.75 million visitors/year.
Only 3.46 billion page views though - Alexa ranks Carnival at 7.1 pages/visit.
That's 658 page views/minute. That would be the baseline they are designed for (11/second).
With 50 ships and weekly cruises, we get 7 ships booking every day. But let's call it 8 to be conservative.
We were estimating
DCL had 250 people for 2 ships, so if the Carnival folk are as dedicated (which I doubt) we're at 1000 people hitting that server at peak. That's 1000 page views/second....
On a system that is normally scaled for 11 views/second.
With DCL we figured a peak of 500x and a conservative minimum of 100x load. With this model we come up with a peak of 90x. What's the conservative minimum? 20x? 40x?
The point is - though it's counter-intuitive, you really are better off with a larger fleet in this regard - because you've already built up more server capacity to handle your average load. Your costs are amortized across more passengers. You've probably also built a more scalable system. Going from 2 servers to 20 can be more complex than going from 40 to 400 (in the latter case you've already solved the database synchronization, load balancing and system management issues).
Also, I do believe that Disboards increases the magnitude of the problem -let's face it, we really do encourage each other (scare each other?) into trying to book at the same time. Does Carnival have anything like this? I've sailed with them a few times and never ran into it. If we had even distributed our reservation load over 24 hours instead of all trying within one hour, I expect DCL would have handled thing quite nicely.