JDK-4771005 : PIORB is slow when ClientRequestInterceptor is used
  • Type: Bug
  • Component: other-libs
  • Sub-Component: corba:idl
  • Affected Version: j2ee1.3.1-fcs,7.0se,1.4.1
  • Priority: P2
  • Status: Closed
  • Resolution: Fixed
  • OS: solaris_2.6,solaris_8
  • CPU: sparc
  • Submitted: 2002-10-30
  • Updated: 2003-01-24
  • Resolved: 2004-12-10
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
Other Other
1.4.1 02Fixed 1.4.2Fixed
Description
Problem Description:
-----------------------
We are seeing extremely poor performance from the PIORB when client 
interceptors are installed. I have a test which does a void-void ping on a 
remote server. On the ORB without any interceptors I can do about 100000 
invokes in 70s. If I install a null interceptor (every method does nothing) 
this drops to 100000 invokes in 200s. Stranger still the results get worse 
the longer I run the test for. So 10000 invokes complete in 10s and then 
increases non-linearly - 20000/21s, 30000/35s, 40000/50s ...

Attached is a testcase. You will need an appserver to deploy the RMI object 
on (or you may be able to do it with the JDK orb. It is dependent on your 
server being quick enough to sustain the load.

unjar and build. Deploy PingerImpl in your appserver.

To run you need to do:

java -classpath . piorbtest.PIORBTest <url> 100000

where url is the address of your CosNaming service that is hosting the RMI 
object.

The test runs for me in 246s. If I remove the call to 
add_client_request_interceptor it runs in 76s. Note that I am not doing 
set_slot or anything like that.

*****************************************************************************

###@###.### 2002-12-10

Fix verified OK.

With build - j2sdk-1_4_2-beta-bin-b09-solaris-sparcv9-04_dec_2002
With test  - tests/performance/interceptors

*****************************************************************************
Comments
CONVERTED DATA BugTraq+ Release Management Values COMMIT TO FIX: 1.4.1_02 mantis FIXED IN: 1.4.1_02 mantis-beta INTEGRATED IN: 1.4.1_02 mantis-b09 mantis-beta VERIFIED IN: mantis-beta
14-06-2004
EVALUATION We have measured the impact of PI in the past and did not see this magnitude of slowdown. ###@###.### 2002-11-19 Using a test produced by Sony I can reproduce the slowdown and see a memory leak. I have found and fixed the problem. ###@###.### 2002-12-20 This is a problem in both the Mantis code base and in the SunOne Application Server code base. This fix plugs a memory leak which occurs in the ORB of the S1AS7.0 container and needs to be fixed in an update release there as well.
11-06-2004
SUGGESTED FIX For the application server portion of the fix, here's the relevant information: The fix was limited to a single file com/sun/corba/ee/internal/Interceptors/SlotTableStack.java. One could check out the diffs in Bonsai at http://cvsync.red.iplanet.com/bonsai/cvsqueryform.cgi. Or Do the diffs manually by using (cvs branch name: iASDev_Branch) : cvs diff -r 1.1.2.3 -r 1.1.2.4 SlotTableStack.java I am attaching the review template and the diffs again for anyone who wants to see: Module: orb Review Template --------------- Bug # : 4771005 Synopsys :orb: PIORB is slow when ClientRequestInterceptor is used CTS Impact : None Perf Impact : Yes UI Impact : None. Doc Impact : None Brief description : (From Harold Carr's mail explaining the fix that he made) SlotTableStack.pushSlotTable uses List.add which causes existing elements to get shifted "right" causing the table to grow without bounds and to hold onto data in those unbounded slots. Also, popSlotTable did not null unused slots so any data they held would be released. This behavior has been corrected. Test Case: This has been tested by the ORB SQE team. The details of the tests can be got from the bug report and from the SQE team. No new tests need to be written to test this on the appserver. Diffs: Index: SlotTableStack.java =================================================================== RCS file: /m/src/iplanet/ias/server/src/java/com/sun/corba/ee/internal/Interceptors/SlotTableStack.java,v retrieving revision 1.1.2.3 diff -r1.1.2.3 SlotTableStack.java 112c112,122 < tableContainer.add( currentIndex, table ); --- > // NOTE: Very important not to always "add" - otherwise a memory leak. > if (currentIndex == tableContainer.size()) { > // Add will cause the table to grow. > tableContainer.add( currentIndex, table ); > } else if (currentIndex > tableContainer.size()) { > throw new org.omg.CORBA.INTERNAL("currentIndex > tableContainer.size(): " + > currentIndex + " > " + tableContainer.size()); > } else { > // Set will override unused slots. > tableContainer.set( currentIndex, table ); > } 115a126 > 135a147 > tableContainer.set( currentIndex, null ); // Do not leak memory.
11-06-2004