Currently when classes are archived by CDS, by default only a small, fixed set of strings are interned at dump time. In order to work better with AOT, it's desirable to intern and archive all the strings. This will allow AOT to produce smaller and faster code.

The CDS archive size will increase approx 2~3%, but this should be more than compensated by savings in AOT. Also, the shared interned strings are effectively read-only (unless the program explicitly synchronizes on them), so their physical memory can be shared across processes.