-
Enhancement
-
Resolution: Duplicate
-
P3
-
None
-
7u4
-
generic
-
os_x
This bug shadows a JIRA bug created for the Mac OS X port at...
http://java.net/jira/browse/MACOSX_PORT-558
Here is the description from that bug:
The Apple Java team modified some of the default VM parameters in jdk6, including the default GC algorithm. The main goal was to improve application startup time and application responsiveness. We did not integrate these changes in the code contributed to jdk7.
The historical underpinnings behind these changes came out of our PPC processor support. We did not have a server compiler (neither 32 nor 64 bit), but we changed VM parameters based upon whether java was invoked with -client or -server.
Upon moving to x86 processors, we provided the 32 and 64-bit server compilers. As we did not have a 64-bit client compiler, we adjusted the VM parameters of the 64-bit server compiler when run in -client mode (leaving them as set by Sun in -server mode). The ultimate goal was to improve desktop startup time. Critical benchmarks were stopwatch time of eclipse and netbeans from command-line initiation to complete initial window draw.
Since apple came late to the x86 world and essentially all macs were dual core with 2gb or more memory, all of our machines fit the traditional Java Server-class definition. We found that CMS worked better than parallel GC for startup time memory use.
Our specific jdk6 findings are not relevant in the light of tiered compilation and jdk7 changes. The purpose of this bug is to get someone at Oracle to benchmark desktop startup times on mac and other platforms and adjust ergo parameters to give the best user experience.
http://java.net/jira/browse/MACOSX_PORT-558
Here is the description from that bug:
The Apple Java team modified some of the default VM parameters in jdk6, including the default GC algorithm. The main goal was to improve application startup time and application responsiveness. We did not integrate these changes in the code contributed to jdk7.
The historical underpinnings behind these changes came out of our PPC processor support. We did not have a server compiler (neither 32 nor 64 bit), but we changed VM parameters based upon whether java was invoked with -client or -server.
Upon moving to x86 processors, we provided the 32 and 64-bit server compilers. As we did not have a 64-bit client compiler, we adjusted the VM parameters of the 64-bit server compiler when run in -client mode (leaving them as set by Sun in -server mode). The ultimate goal was to improve desktop startup time. Critical benchmarks were stopwatch time of eclipse and netbeans from command-line initiation to complete initial window draw.
Since apple came late to the x86 world and essentially all macs were dual core with 2gb or more memory, all of our machines fit the traditional Java Server-class definition. We found that CMS worked better than parallel GC for startup time memory use.
Our specific jdk6 findings are not relevant in the light of tiered compilation and jdk7 changes. The purpose of this bug is to get someone at Oracle to benchmark desktop startup times on mac and other platforms and adjust ergo parameters to give the best user experience.
- duplicates
-
-
- Closed
-