Glancing through the cheetah book, there's three ways to get data from DBI: row methods (each time you call the method, it returns the next row), atomic methods (for returning only one row, doesn't help you), and batch methods (fetchall_arrayref(), selectall_arrayref(), like chromatic mentioned). The batch methods can indeed knock a system on its behind if the result set is big; it's certainly better to prepare and execute your SELECT, roughly make sure you can handle the results, then use fetchall_arrayref rather than selectall_arrayref. If you don't use all the data, though, this can be wasteful. If you're grabbing sequential sections of the result set (which it looked like from your question), I don't see a problem with just doing atomic fetches 20 times a pop when you need the next set.

If you're using MySQL (I haven't used any other rdbms), LIMIT is a good option, like others have said. From the manual: "mysql> select * from table LIMIT 5,10; # Retrieve rows 6-15" looks like what you need. But unless the offset parameter of LIMIT can be a bound value (anyone know?), I think you'd take a performance hit for having the DB redo the query plan each time.

A side note that bit me: if you want to use LIMIT with UPDATE, you'll need MySQL 3.23.