I created some simple CMP test beans and found that 1.1 is just plain
slower than 1.0. The test beans are:
long pk, long field - a zero string test bean
long pk, string - a one string test bean
...
long pk, 5 strings - a five string test bean
The test program is also simple - create(pk), set all fields, get all
fields in a loop, followed by a loop that removes all the beans. I ran
this on both 1.1.37 and 1.0 (not 1.0.3b), changing only orion.jar between
test sessions to rule out other issues, and got the following elapsed
times. These are ms/pass for 1000 passes, and repeatable:
1.1.37 1.0
create/set/get remove create/set/get remove
Zero 17 13 8 4
One 17 12 8 5
Two 23 13 11 4
Three 29 12 15 4
Four 35 12 18 5
Five 41 12 22 5
1.1.37 runs 2x slower than 1.0 in this sample. The beans and test programs
come out of a production code generator, and the test was intended to do
some bean verification rather than checking performance. These results are
consistent with my earlier hand-coded BMP tests, which also ran
consistently slower, and with similar runs against our production code's beans.
My test platform is still rather crude (more and better hardware is on
order...), a P3-550/128MB/Win98/JDK 1.2.2/Hypersonic (in server mode, same
machine (thanks for the example!)).
Please, some explanation. Is this a bug, debug code in Orion, or a trend
for future versions? We're getting concerned, particularly since we don't
have any explanation or feedback.
Kirk Yarina
[EMAIL PROTECTED]
