JDK-6646599 : Classloading in 1.6.0_03-b05 server VM is much slower than 1.5 server VM
  • Type: Bug
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 6
  • Priority: P3
  • Status: Closed
  • Resolution: Duplicate
  • OS: windows_xp
  • CPU: x86
  • Submitted: 2008-01-02
  • Updated: 2011-02-16
  • Resolved: 2008-10-22
Related Reports
Duplicate :  
Description
FULL PRODUCT VERSION :


A DESCRIPTION OF THE PROBLEM :
Startup in 1.6.0_03-b05 server VM is much slower than 1.5 server VM. It mainly looks like classloading problem. I took the numbers for my sevrer application which needs to load huge number of classes (> 10K applications classes) at startup.

With nearly 10K classes, the times reuired to load these classes were
1.5 client VM : 64 sec
1.5 server VM : 27.5 sec
1.6 client VM : 66 sec
1.6 server VM : 53 sec

With 15K classes, the difference became even more
1.5 client VM : 236 sec
1.5 server VM : 128 sec
1.6 client VM : 274 sec
1.6 server VM : 267 sec

1.6 server VM performs nearly 50% slower than 1.5 server VM. In fact as you keep on increasing the number of classes to be loaded, the performance keeps on degrading.

Seems many people have observed this :
http://forums.java.net/jive/thread.jspa?messageID=204324
http://forum.java.sun.com/thread.jspa?threadID=5218663

In some cases we are seeing that complete initialization of application takes nearly 20 minutes on 1.6 server VM whereas it takes less than a minute on 1.5

Our customers can not move to 1.6 unless this is resolved.


REPRODUCIBILITY :
This bug can be reproduced always.

Release Regression From : 6
The above release value was the last known release where this 
bug was not reproducible. Since then there has been a regression.

Comments
EVALUATION No feedback from cutomer in 6 months. Closing as a duplicate.
22-10-2008

EVALUATION This bug is probably a duplicate of 6471009, though I cannot verify this without the customer's regression test case. The fix for 6471009 is now in the 6u10 beta (6u10-b09 and later) and the 7 builds (7-b20 and later). I will close this as a duplicate bug unless there is information that the regression is still observable in 6u10.
23-04-2008

EVALUATION Without a test case, it is only conjecture that this bug is the same as 6471009. Assigning John Rose as RE, who may close this one as a duplicate of that one.
07-01-2008

EVALUATION Likely due to 6471009 which has been fixed in jdk7. I believe the solution to that bug is going to be in one of the jdk6 update releases. Routing to the VM team for verification that this is the same problem and that they plan to backport 6471009 to an update release.
03-01-2008