Hi Alan,
thanks for your interest, and your hints.
- Yes, I'm on Windows.
- I did my tests in default (-client) mode with 100 loops on 190k file.
In my experience I didn't recognize any significant difference for 500
or 1000 loops, so IMO 100 loops should be enough. On the other hand I
had significant differences (up to 3 times) on different runs of
SpeedTest at different times, but regardless of 100 or 1000 loops. I
assume, that other running windows tasks are the cause for this. So I
did up to 40 runs, and then valued the minimum of those results.
- Now I also tried some tests with the -server option. The improvement
is ~10 % for 100 loops, but ~30 % for 1000 loops.
-Ulf
Am 19.05.2008 13:16, Alan Bateman schrieb:
Your scripts are DOS batch files so I assume you are on Windows. As
the scripts don't specify the -client or -server option then I assume
you are testing only with the Client VM (is that right?). If you are
only testing with the Client VM then switching to the Server VM should
make a significant difference to this test. Another comment is that
the input file is only 190k so you'll probably need to decode it
several thousand times to get to a steady-state and get reasonable
results.
One other thing - it looks like you are using the jdk6 sources with
the JRL license rather than OpenJDK sources. If you do improve the
performance of the decode loop for this class of decoder then it would
be best to post a patch against the OpenJDK sources.
-Alan.
