keeping thousands of unused akas in case any of it might be useful for its ip at one point? in what universe would an IP match with a 0-0-0 account from 2 years ago amount to anything?
Let me try to explain:
Deletion is not a term in Big Data. Archived is however an option.
It's just a term, not the only guide to action.
I also prefer to archive data, not to delete it. I mean data could be useful for investigation for future. BNET record data is.
Example: you can definitely know several cheaters (like dellam) create tons of accounts every day. Collecting that history today can help to find some patterns for future. It's not only ip addresses, it's also account creation time, last login time, etc.
It's just an example applied to Big Data philosophy. It may be not our case: i just deleted about 80k accounts and don't care of that. Because it was the easiest way to free the accounts from the database and decrease the database size.
And i definitely know i will never have time to make any smart investigation patterns for them.
But as for serious example, like social networks - i'm pretty sure they never delete any records about users once created in their databases...