5613 2007-10-12 16:37 Profile Java WS Core container/dispatch code base 2008-09-02 10:03:24 1 1 1 Unclassified Java WS Core Campaign unspecified PC Windows XP RESOLVED FIXED P3 normal 4.0.6 5653 5665 5612 1 ranantha@mcs.anl.gov trhowe@uchicago.edu jwscore-dev@globus.org rohder@mcs.anl.gov ranantha@mcs.anl.gov 2007-10-12 16:37:52 Definition: This campaign will be the first of a set of campaigns to analyze and improve reliability and scalability of the core container and dispatch functionality. A test suite to exercise concurrent access of the container will be developed and the code will be profiled to identify any issues. There will be follow-up campaigns to analyze and implement solutions to fix the identified problems. Benefits: Preliminary scalability testing of GRAM indicates that the core container is limited in the number of concurrent requests it can handle, before failing over. This campaign should help identify the issues and provide input for follow-up campaigns to fix them. This improve scalability in the container and directly impact GRAM and other higher level services built on top of Java WS Core. Also, the test suite and methodology developed will provide a tool for periodic performance measurements in the future. Roadmap : http://bugzilla.globus.org/bugzilla/show_bug.cgi?id=5612 Deliverables: - Design and implement the following: - Test service with a single operation - Test client that invokes the operation either concurrently or sequentially, based on configuration - Profile Java WS core code, using the above tests. - Identify tools to use - Specific focus on the container and dispatch code. - Document issues identified. ranantha@mcs.anl.gov 2007-10-12 16:43:45 I missed adding that this work needs to be on the 4.0.x branch. trhowe@uchicago.edu 2007-11-07 11:57:23 I think this problem may be connected to: http://bugzilla.globus.org/globus/show_bug.cgi?id=3449 ranantha@mcs.anl.gov 2007-11-13 12:05:48 Status update based on discussion with Tom: Developer notes: http://dev.globus.org/wiki/Java_WS_Core/Development_Documentation/Profiling_Container Two bugs that were identified with initial profiling have been fixed. Tom has rerun the profiling and he does not see the thread contention issues from before. Next set of tests that are being profiled are for usage of threads from core thread pool and GRAM code for thread pooling (different from core). Test 4 and Test 5 in http://dev.globus.org/wiki/Java_WS_Core/Development_Documentation/Profiling_Container ranantha@mcs.anl.gov 2007-11-26 11:25:35 Committed patches provided by Tom in cryptix-asn1 and puretl repository. The patch has been applied to Tbug_5613_3 branch and the updated jars have been committed to wsrf repository in the branch. trhowe@uchicago.edu 2008-03-28 14:28:30 The problems discovered here were related to a bug in the way axis handled one-way operations. Each notification required a new thread be created. As a result, when many notifications were submitted simultaneously, too many threads were created causing the container to choke. A patch was submitted. trhowe@uchicago.edu 2008-03-31 15:23:50 Summary updated at: http://www-unix.mcs.anl.gov/~trhowe/profiling/index.html trhowe@uchicago.edu 2008-05-01 07:50:13 Updated results at: http://dev.globus.org/wiki/Java_WS_Core/Development_Documentation/Profiling_Container trhowe@uchicago.edu 2008-09-02 10:03:24 Finished profiling 4.0 container