Uniquely identifying files

I have a whole bunch of files I need to track in a database (and create associated data on). The filenames will be static (but not necessarily unique), and the contents of the files are unique, but the problem is that the directories they are in are not necessarily going to be staticly named. Originally I was thinking of storing directory/filename into a db and uniquely identifying by that, but the directory names will be changing. I was thinking of storing an md5 hash of say the first 2k of every file into a database, and comparing the hash for a file retrieved with the md5 hashes in a database and retrieving data associated with it. My only problem is this is in theory going to be processor intensive. Has anyone got any other suggestions on how to do this. I'm kind of adverse to renaming the files uniquely, as the filenames are meaningful, My current thoughts are MD5 or CRC32 on the file contents, any ideas would be appreciated. Thanks Lawrence.