In memory data structure can be sizeable if the on-disk cache is large

[rpetrich]

I've switched to using SQLite in my fork. There is no migration code, but it could be useful as a starting point:
https://github.com/rpetrich/SDURLCache/tree/sqlite

[rs]

Maybe the index storage logic could be split into another class so we could make them plugable and easier to maintain?

[rpetrich]

I am in favor of that.

[rs]

And what about another storage backend more memory efficient and faster than SQLite like for instance Kyoto Cabinet or LevelDB?

[rpetrich]

I've found SQLite to be sufficiently memory efficient with the default settings and sufficiently speedy with SQLITE_OPEN_NOMUTEX and PRAGMA synchronous = 1;

Would be interested to compare with other data stores though. If you make it pluggable, I would be willing to commit and possibly maintain the SQLite backend.