Active users, not the total number of users in the user table. People grossly overestimate the scale of most web properties, where the number of active users is far lower than you likely imagine.
>But a bunch of industries have opened up within the last ten years that require several orders of magnitude more data, and it's naive to think that just because it works for a corporate help desk or POS system, it'll work for a system that logs every page view and every action for millions of users.
Strangely it says nothing of the sort. Yet here, again, you've used Google as the example. How many Googles are there? How much does that apply to about 99.999% of people who deal with databases?
Yet it always appears as the example.
There's tens of thousands of alternative search engines out there, see this for a sample from '07:
http://www.altsearchengines.com/2007/10/29/the-top-10-list-o...
Most of them fail, but there's a lot of people trying to solve Big Data problems on a shoe string budget. That's part of the disconnect, us NoSQL folks are excited by having cheap solutions to problems we wouldn't be able to afford to tackle otherwise, whilst SQL folks are are shaking their heads at the mess we'll have to clean up if our prototypes do become successful.
Say that his internal help ticket tracking system was built for IBM, one of the largest corporations out there with 300,000 employees. 300k users is tiny for a consumer app. We had more than that when I was volunteering for a Harry Potter fanfiction website. Even if he was working at the largest company on earth by employee count (Wal-Mart), he'd still have fewer users than we had for Harry Potter fanfiction. And usually employees don't submit more than one or two help tickets a day, while Harry Potter fans tend to view a forum thread every minute or so.
It really hits home how consumer data processing has changed the game for data management. When I was working in the financial industry, we dealt with about 50GB of data/day coming off the exchanges. I thought that was a lot. But at Google, there's terabytes per day - at least two orders of magnitude more - and the total volume of financial transactions is basically rounding error on the data we handle.
It makes sense that with this exponential explosion in data, we'd need different techniques to handle it. Quite likely, RDBMSes do scale for the scale he's talking about. But a bunch of industries have opened up within the last ten years that require several orders of magnitude more data, and it's naive to think that just because it works for a corporate help desk or POS system, it'll work for a system that logs every page view and every action for millions of users.