JDK-6381820 : Performance regression observed for DaCapo chart benchmark
  • Type: Bug
  • Component: core-libs
  • Sub-Component: java.util.jar
  • Affected Version: 6
  • Priority: P3
  • Status: Closed
  • Resolution: Duplicate
  • OS: generic
  • CPU: generic
  • Submitted: 2006-02-06
  • Updated: 2011-12-15
  • Resolved: 2006-03-17
Related Reports
Duplicate :  
JDK-6364346 - GZIPOutputStream is slower on 1.4.2_11-b02 than on 1.4.2_09
Relates :  
JDK-6379969 - Performance regression observed for DaCapo xalan benchmark
Description
CAP Bug Submission Form
Please fill out the following  as completely as possible and email to
###@###.###.  Please also include any source code or class files
required to reproduce the bug.


Your Name: Monica Bhan-Beckwith	
Company Name: AMD


J2SE Version (please include all output from java -version flag):
   Java HotSpot(TM) Server VM (build 1.6.0-rc-b67 mixed mode)

Does this problem occur on J2SE 1.4.x or 5.0.x ?  Yes / No (pick one) 
   No


Operating System Configuration Information (be specific): 
   SLES9 (64bit) (2.6.5-7.97)

Hardware Configuration Information (be specific):
   V40Z system, Opteron 875 - 4#, 16Gs of memory.


Bug Description: Bug for: java/hotspot/compiler2
   Performance regression observed for 2 benchmarks (chart and xalan) 
   in the DaCapo suite when compared to 1.6.0-ea-b56

Steps to Reproduce (be specific):
   Download the suite from  - http://www-ali.cs.umass.edu/DaCapo/gcbm/dacapo-beta051009.jar

    Run with the 2 JVMs (1.6.0-ea-b56 and 1.6.0-rc-b67) for both 32 and 
    64 bit versions. You will observe the regression for chart and xalan
Comments
EVALUATION The fix for 6364346 improves the performance of deflation. The performance of chart is back to where it was; that of xalan is unchanged. Timing results below (other output omitted here for clarity). On dual AMD64, SunOS 5.11: % jver 1.6.0-b64 java -cp . -jar dacapo-beta051009.jar -two chart ===== DaCapo chart Finished in 6986 msec ===== % jver 1.6.0-b65 java -cp . -jar dacapo-beta051009.jar -two chart ===== DaCapo chart Finished in 9517 msec ===== The fix in this workspace has been putback to integration workspace. % /data/ws/jarzip/build/solaris-i586/bin/java -cp . -jar dacapo-beta051009.jar -two chart ===== DaCapo chart Finished in 6980 msec ===== With the fix to 6364346 in place, the suggested fix (to increase the default buffer size in DeflaterOutputStream) does not make a noticeable difference.
17-03-2006
SUGGESTED FIX The size of the buffer used by DeflaterOutputStream (and similar in java.util.zip) has been 512 bytes since 1996. Changing that to a more modern 4096 or 8192 might resolve the problem.
07-02-2006
EVALUATION Looks like a dup of the problem introduced by the fix for 6206933: GZipOutputStream/InputStream goes critical(calls JNI_Get*Critical) and causes slowness Is it possible for the classes that wrap inflaters and deflaters to automagically do the buffering we expect users to do?
06-02-2006