Open ghost opened 14 years ago
I'd like to learn more about this use case because AFAIK BDB-JE runs at least at disk speeds when doing fully transactional writes due to its log-based append-only on-disk structure. When buffering commits it can be much faster. In addition, the most recent version of BDB-JE has been updated to run much faster when doing random reads. (Disclaimer: I am a product manager in the Berkeley DB Group at Oracle). Please contact us via the BDB JE OTN forums (http://forums.oracle.com/forums/forum.jspa?forumID=273) so we can help you use BDB as the default storage layer for your project.
It would be more convenient if we could support Berkeley DB Java edition by default. That would eliminate the need for including multiple native code libraries for BDB in the release. To do this, though, the performance of the Java edition needs to match that of the standard edition. So far, although the Java edition's object-level locking is better for avoiding conflicts between reads and writes to objects, the Java edition's lock-next-object scheme produces bad concurrency conflicts when creating objects. Need to track down these performance problems.