>We were doing a load test on a new system here and needed to generate 10 million name and address records that had "statistically valid" distributions for name and address information to make sure index fragmentation would be the same as it would be when the system went into production. I used real data from another couple of systems to create probability distributions for elements like first name, lastname, address, city, state and zipcode. These distributions were used to then generate "realistic" random data.

I haven't worried too much about statistical validity, but actually the model I'm using could handle it without much tweaking. My goal is to create realistic volumes of realistic data, instead of always testing with three recods that contain John Smith of 123 N. Main St. <g>

Tamar