Testing your JavaFX applications on JDK 9

[Scott Palmer]

I’ve only tested with Java 9 on Windows 10 64-bit so far.

I have not yet done any profiling to try to isolate where the difference might be.  The application generates data in native code, wrapped in a native ByteBuffer and passes an object around holding the buffer and map of properties to go with it.  I’m doing virtually no processing on the data and just measuring the rate that packets are going through the workflow so far.  The GUI is mostly static during the test except for a table and some labels showing stats that are updated with bitrates, packet rates, timestamps etc.

I’ll continue to try to isolate a test case that I can share as time allows.  I should be able to get some testing done on Mac and Linux as well and will send a followup email with any new information.

Regards,

Scott


> On Jan 26, 2017, at 9:37 PM, Philip Race <philip.race at oracle.com> wrote:
> 
> Scott,
> 
> What VM/bitness/platforms show this ? All or just a specific one ?
> 
> -phil.
> 
> On 1/26/17, 6:08 PM, Kevin Rushforth wrote:
>> Hi Scott,
>> 
>> Thanks for the report. The only JDK 9 specific performance problems I have seen is with startup, which would be a good thing to discuss on jigsaw-dev. If you are seeing actual application slowdown, that might be something else, depending on where the problem is. Have you profiled your application?
>> 
>> -- Kevin
>> 
>> 
>> Scott Palmer wrote:
>>> I finally managed to get my app running on JDK 9.  It is based on a plugin architecture and the plugins are made accessible by modifying the system class loader at runtime (because that seemed to be the only way to make things like JAXB able to reach the classes in the plugins).  That “hack” broke with the module system so I had to find an alternative.  (Now I’m using a java agent to get an Instrumentation interface so I can do it in a somewhat supported way via the appendToSystemClassLoaderSearch method.)
>>> 
>>> Anyway, I’ve noticed that the application runs slower on Java 9.  For a simple test I see almost a 10% performance drop.  However, I’m not sure that it is Java FX related.  I see the same throughput with the default renderer or the recently added Marlin renderer, always a bit slower than Java 8u121.  I use the same JVM options, including forcing the garbage collector to use the same algorithm.
>>> 
>>> Is there any expectation of a performance drop with JDK 9 at this stage?  For example should I expect some unoptimized stuff in the EA builds?
>>> I guess the jigsaw-dev alias the right place to bring this up?
>>> 
>>> I have a feeling it is going to be difficult to track down. :-(
>>> 
>>> I have noticed a couple layout issues fixed with JavaFX 9. Stuff that used to “snap into place” only when I resized a dialog now seems to be correct when the dialog first appears. Thanks for that!
>>> 
>>> Regards,
>>> 
>>> Scott
>>> 
>>> 
>>>> On Jan 11, 2017, at 2:14 PM, Kevin Rushforth <kevin.rushforth at oracle.com> wrote:
>>>> 
>>>> As you may know JDK 9 has hit the feature extension complete milestone [1]. We still have a small number of weeks to fix P1-P3 bugs, but we need to know about them in order to fix them. Our focus will be on bugs that are new in JDK 9 (regressions).
>>>> 
>>>> I ask all JavaFX developers to please download JDK 9 early access [2] and test your application. You can report any bugs via bugs.java.com [3]. If you have any questions about whether a behavior is a bug or is expected, please ask on this alias or on the jigsaw-dev alias [4].
>>>> 
>>>> Thank you.
>>>> 
>>>> -- Kevin
>>>> 
>>>> [1] http://openjdk.java.net/projects/jdk9/
>>>> [2] https://jdk9.java.net/download/
>>>> [3] http://bugreport.java.com/
>>>> [4] jigsaw-dev at openjdk.java.net
>>>> 
>>>