IJ53585: REDUCED MD5 PERFORMANCE

APAR status

• Closed as program error.

Error description

• Error Message: Increased cpu time was noted in MD5 codepaths
  after updating from Java 8 SR8 FP25 to FP26 when using balanced
  GC.
  .
  Stack Trace: 3XMTHREADINFO3           Java callstack:
  4XESTACKTRACE                at
  com/ibm/crypto/provider/MD5.a(Bytecode PC:192(Compiled Code))
  4XESTACKTRACE                at
  com/ibm/crypto/provider/MD5.engineDigest(Bytecode PC:17(Compiled
  Code))
  4XESTACKTRACE                at
  com/ibm/crypto/provider/PBEWithMD5AndTripleDESCipher.engineInit(
  Bytecode PC:370(Compiled Code))
  4XESTACKTRACE                at
  javax/crypto/Cipher.init(Bytecode PC:35(Compiled Code))
  4XESTACKTRACE                at
  javax/crypto/Cipher.init(Bytecode PC:7(Compiled Code))
  4XESTACKTRACE                at
  com/ibm/crypto/provider/I.a(Bytecode PC:114(Compiled Code))
  4XESTACKTRACE                at
  com/ibm/crypto/provider/JceKeyStore.engineSetKeyEntry(Bytecode
  PC:184(Compiled Code))
  .
  Other callstacks are possible, but note that the problem has
  only been seen with the IBMJCE version of MD5 and specifically
  impacts com/ibm/crypto/provider/MD5.a. It is common for the
  IBMJCEPlus provider to be used for MD5, with the notable
  exception of the use in JceKeyStore.
  

Local fix

• The problem can be avoided by disabling profiling for the
  affected method, e.g. with the command line option:
  "-Xjit:{com/ibm/crypto/provider/MD5.a(*}(disableProfiling)"
  Note that if there is already an Xjit option being used then the
  above will need to be merged with the existing content.
  

Problem summary

• The problem happens when the JIT is unable to generate a
  profiling version of the MD5.a method, after which the method
  will remain running in the version that was compiled at the
  'warm' optimization level instead of advancing to the
  'scorching' level. The complexity of the method is increased
  both by the code needed to support the balanced gc policy and by
  adding profiling data collection points, resulting in reaching a
  limit on the number of internal pointers.
  

Problem conclusion

• Three changes were made to the JIT to resolve this class of
  problems.
  1. The number of internal pointers added by profiling was
  reduced
  2. In the event of running out of internal pointers during
  profiling the compilation is allowed to complete with partial
  profiling
  3. If a profiling compilation fails for any reason then a
  non-profiled compilation at a high optimization level is
  scheduled to replace it
  https://github.com/eclipse-openj9/openj9/pull/21053
  https://github.com/eclipse-omr/omr/pull/7637
  https://github.com/eclipse-openj9/openj9/pull/21056
  .
  This APAR will be fixed in the following Releases:
  .
  IBM Semeru Runtimes
  IBM SDK, Java Technology Edition
     8    SR8 FP45  (8.0.8.45)
  .
  Downloads and supplementary documentation can be found at the
  following locations:
  - For non z/OS operating systems:
    - IBM Semeru Runtimes, Version 11 and later
      https://www.ibm.com/semeru-runtimes/downloads/
    - IBM SDK, Java Technology Edition, Version 8
      https://www.ibm.com/support/pages/java-sdk-downloads/
  - For the z/OS operating system:
    - Java SDK Products on z/OS
      https://www.ibm.com/support/pages/java-sdk-products-zos
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  JIT

• Fixed component ID

  620700124

Applicable component levels