Particularly when working with large data sets, if the hash can be stored in memory, it really, really should be due to performance considerations.

Secondly if you have to use a DBM file, I highly recommend not using dbmopen. That is because of the following gotcha. You wrote a program that stores data using dbmopen. Your program has been successfully running, storing, and accessing data, every night for months. Then it suddenly stops working one day, and nobody knows how to get at your data. How are you going to figure out that this is because a sysadmin installed DB_File on your machine? And once you do figure it out, how are you going to fix your program?