Performance patch for AIX system api support of sun.nio.ch.FileChannelImpl.transferTo0()

[Alan Bateman]

Charles Lee wrote:
> Thanks Alan. Any requirements for performance patches to contribute? 
> Test case? Test result?
Speaking generally, when working on a performance fix (or any fix for 
that matter) it's always good to understand the test coverage for the 
area that you are changing. The math area is a good example where 
performance work in 7 required extending the test coverage to give 
confidence that performance work didn't cause any correctness issues. 
There are examples in other areas too. When fixing bugs then the rule is 
that tests should be added (or existing tests updated) to cover the 
scenario that is being fixed. However there are cases, including 
performance, where it's not feasible or doesn't make sense to write an 
automated test. Tests need to run quickly so we wouldn't check in a test 
that takes an hour to demonstrate a bug for example. Also it's worth 
keeping existing tests in mind to avoid duplicating existing tests. With 
performance tests then it's as I was saying in the first mail, the jdk 
repository isn't really the best place for them in my view. It's fine to 
push tests that print results that we can run manually in a controlled 
environment (we have several examples of this) but we can't have tests 
that fail if they think that the performance isn't as expected. At some 
point it would it great if OpenJDK could get a repository and framework 
for micro benchmarks and maybe larger performance tests but that's a 
bigger discussion and not really something for this list.

-Alan.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/nio-dev/attachments/20111020/e4d9ef47/attachment.html