Comments inline.
At 10:32 27.03.2002 -0800, Post, Richard (HQP) wrote:
>Hi,
>
>I have a question about the LogTest.java file that you used for your
>performance measurements.
>
>I noticed that the 'log.join()' statement at line 50 is outside the 'for'
>loop (lines 44-47) that creates the threads for each benchmark.
>
>Doesn't the 'log.join()' statement cause the code to wait until the 'log'
>thread dies before continuing?
>
>If so, then this code will wait until the last thread created in the 'for'
>loop (44-47) dies; this is because each thread created is used to populate
>the 'log' variable. Thus when the 'for' loop ends the 'log' variable will
>represent the last (3rd in this case) thread created.
>
>There is nothing to guarantee that the last thread created will be the
>last thread to die. I have run tests where the last thread created is
>often the first thread to die; it acts like the last thread monopolizes
>the PC (NT Workstation in my case) and runs to completion.
>
>If the last thread is not the last to die then the code in LogTest.java
>that calculates the 'msec:' and 'Logs/sec:' times will not necessarily
>include all of the events.
Very good point.
>Just a couple of side notes:
>
>1) Since the value '15001*1000.0' in line 56 is a constant the value could
>be stored in a variable prior to both 'for' loops and not calculated each time.
I think the compiler will do it for you. Anyway, it should not matter that
much.
>2) It does not appear that the 'logger' statements in the 'run' method in
>the 'LogTester' class make any distinction as to which thread they are
>being logged from. It might be helpful to 'tag' each statement with the
>thread id to make it easier to verify which entries in the log file where
>logged by which iteration of which thread.
The conversion pattern is:
log4j.appender.A1.layout.ConversionPattern=%d{ISO8601} %-5p [%t] %c{3} %x -
%m%n
So the output should be tagged. On the other hand, it'll be cool to measure
the impact
of NDCs, MDCs and Localication info.
>Good benchmark results though!
Definitely. I wonder if on NT the numbers are similar. Ceki
ps: By the way, before I forget, thank you Ole Dalgaard.
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
