Using Devel::Size for this is likely to cause problems because it uses a fairly substantial amount of memory itself when calculating the size of nested data structures. Much in the same way that Data::Dumper does, it uses it to eliminate cyclical references.

I'd advise just using the cachesize parameter (via BerkeleyDB::Env class ) and set it as large as possible without pushing the process into swapping. It will probably do a better job of balancing performance than you can.