Berkeley DB Shortcomings
Posted 1:41 PM

After trying exhaustively to get Berkeley DB to stand up in a production environment, we switched to "TDB" as our embedded database instead.

Berkeley DB just doesn't deal well with processes and threads dropping without cleaning up nicely. In a heterogeneous/complex environment, good database systems need to respond very well to dropped connections. Perhaps there's a way to coax BDB into timing out locks more quickly or allocating more locks, but I couldn't get it to work.

TDB was able to handle a very high load (thousands of reads/writes per second) while also having 1 in 100 connections get dropped or killed without being gracefully closed.

Perhaps it's just a matter of TDB having better defaults for that sort of scenario.
[View/Post Comments] [Digg] [Del.icio.us] [Stumble]