While i agree with blandry that having a solid idea of data throughput is nearly impossible, i don't think that invalidates Phil's point. I think what Shaun was talking about is the perfect example:
Shaun McGuile (12/18/2008)
In my case the 3 million rows was full population of the database for six months worth of usage.I had specifc fixed limits in the system I was testing and actual data would be at most 2 million rows, given the nature of the application. So doing all the load testing with 4 - 6 times the allowed data volumes was a good test.
hey thats the beauty of redgate sql data generator though. 😉
He had an inkling of what production numbers would look like, but the more important thing was performing the test itself. If you can't know how much data is going to be in the database (and you have data generator...), filling it up with a ludicrous number of rows and doing performance tests is still valid.