faster and faster

Well, I've done some tuning now.

• Null transactions (e.g. queries): 10,000,000/sec
• Aborts: 4,000+/sec
• Small transactions: 100+/sec

To get this, I've stopped zipping the data and only calculate checksums on the root. I'm also using rwd as the mode on the random access file instead of doing a sync when writting the root.

I'm also looking at fully implementing the copy-on-write technology, which allows you to create snapshots of the current state of the database with virtually no overhead. I see two major advantages here:

• Backups can be done while updating the database and
• Queries can run while updating the database.

Both of these are significant. The first allows a database to be up 24x7. The second means that the response time for queries will always be low.