Fairly often I hear customers say that they plan a table that accumulates millions of rows per day, and they want to keep around, say, the last 30 days worth of data. (For the sake of examples, I'm going to make it the last 3 days.) So this is a kind of round-robin table, with rolling addition of new data and removal of the oldest data.

With a high volume of data, this sounds like a table partitioned on day boundaries (in MySQL 5.1). See Sarah's blog and her links for a quick ramp-up on time-based table partitioning (http://everythingmysql.ning.com/profiles/blogs/partitioning-by-dates-the). One great benefit of table partitioning is that you can drop a partition to lose millions of rows in one quick statement, much faster than deleting millions of rows. Sort of like a partial TRUNCATE TABLE.

First create the table with 4 partitions, and then, once a day, drop the oldest partition and add another partition to store the …