Bill Wagner wrote:
>Time of execution for StringBuffer test:
>In the Mac with VM 1.3
> 20 ms
>Same execution now with VM 1.4.2_12:
> 17 ms
>Using VM 1.5.0_07:
> 26 ms
>
>In 1.3, we get 20ms, then in 1.4 it improves to 17ms, then in 1.5 the
>execution time goes up to 26ms, which means that processing time using 1.5
>has degraded by slightly over 50% with respect to 1.4.
An alternative interpretation...
You have 3 data-points with an average of 21 ms, and a variation of about
+/- 4 ms. This seems well within the normal bounds of run-to-run variation.
We don't know how many times the test was run to produce each data-point,
nor do we know what the length of each test-run was. We don't even know
exactly what the test was. The DDJ article mentions run times of 100's ms,
not 20 ms, so either the article's test was run on a much slower machine or
the test itself was changed substantially from the one in the article.
Any time you measure Java performance you have to account for the JIT
compiler. When the JITC kicks in, it produces a sharp "knee" or step in
the performance profile. If you don't account for that, it's impossible to
draw any useful conclusions about overall performance. And on some CPUs
you also have the -client vs. -server HotSpot compilers, which have notably
different performance profiles. Search the list archives for HotSpot for
past posts on performance.
>It would seem to me that this indicates those performance improvements
>have not been implemented in 1.5, which is an official release now.
I'm reluctant to draw any specific conclusions about overall performance
from so few data-points, especially with so much that's either unstated or
uncertain. This doesn't mean there isn't a measurable performance
difference, just that the evidence presented so far is anything but
conclusive. Consider it an open invitation to perform more extensive and
rigorous tests. The best way to show there's a performance issue is with
clear tests, consistent results, and enough data-points to conclude
something useful.
However, since I think the main point of the initial post was the outlier
136 ms reading, removing it from consideration leaves only a teacup-sized
tempest, which may or may not turn into a more substantial weather system.
-- GG
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Java-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/java-dev/email@hidden
This email sent to email@hidden
