On Mon, Jun 21, 2010 at 04:47:12AM -0700, Igor Sereda wrote: > > Wow, that's interesting news. Berkeley DB is still GPL/commercial, I guess? I > hope SQLite will keep on going under public domain, including its B-tree > level. > > Also, here's an interesting statement in the BDB/SQLite announcement: > > > > Thus, applications written to the SQLite version 3 API can switch to using > > Oracle Berkeley DB with no code changes, by re-linking against the > > Berkeley DB SQLite library. The application-level behavior of the two > > products is identical, but the advanced features of Berkeley DB provide > > SQLite applications improved performance, concurrency, scalability, and > > reliability. > > > > Could you please comment on that? > Does that mean SQLite storage level is less reliable than BDB?
<disclaimer> My own opinion, not that of DRH </disclaimer> If you look at the features of BDB, it includes functionality such as replication, enabling high availability across multiple hosts. From an integrity point of view, SQLite is generally as reliable as the hardware underneath it. From an outside point of view, SQLite is only as reliable as the box it is running on, and should the box be unavailable, so is your SQLite data. > Are there any performance measurements and comparison of SQLite vs. > SQLite/BDB? > Does SQLite/BDB really provide more concurrency, while maintaining > SERIALIZABLE isolation level? BDB provides MVCC, though I've no idea if this spans process boundaries. MVCC is well known for allowing increased concurrency but is difficult to implement without extra IPC primitives that SQLite doesn't have access to. That said, the WAL work currently in progress appears to offer simple MVCC conceptually similar to how Oracle implements MVCC (keeping old copies of pages about), so this should increase concurrency somewhat in the cases where WAL can be used. BDB can probably enable multiple writers if it uses versioned rows or page/row level MVCC, as opposed to SQLite's table level locking. I'd be surprised if Oracle post any SQLite vs. SQLite/DBD benchmarks, and I'd also be surprised if BDB licensees are allowed to publish such data. Regards, Christian _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
