Cursor spending too much time cloning

Question

I am profiling the performance of scanning through a database using a cursor. The breakdown looks something like this:
Total - 3800 ms
IN.fetchTarget - 1250 ms
IN.latch - 300 ms
Cursor.beginMoveCursor - 1260 ms
Cursor.endMoveCursor - 920 ms
Most of the time in begin and end move cursor is dealing with cloning. So effectively it is taking more time to deal with the cursor than to retrieve the data from disk. That doesn't sound right. Am I doing something wrong? Is there a way to not clone the cursor?
Instead of using the cursor if I use get for each element the time is comparable. Here is the breakdown
Total - 4800 ms
IN.fetchTarget - 1250 ms
Tree.search - 940 ms
Cursor.init - 1380 ms (almost all the time incrementing an atomic integer)
Cursor.close - 830 ms
Cursor.lockLN - 300 ms
I was hoping that by using the cursor I will cut down most of the overhead (search, repeated cursor init and close) and the time will come down to close to 1250 ms.

Answer 1

I strongly suggest that rather than doing profiling, you look at your app throughput (without -ea or a profiler running), and look at the EnvironmentStats. The stats tell you what is going on in JE, and tuning based on the stats is normally the most fruitful. Looking at things in a profiler may help to optimize JE code itself, but is less likely to be fruitful for tuning your app, and a profiler is usually more beneficial after you have tuned the app, based on looking at the stats.
--mark