JDK-8209538 : Micros-Crypto performance regressions in 12+6 need investigation
  • Type: Bug
  • Component: performance
  • Sub-Component: libraries
  • Priority: P3
  • Status: Closed
  • Resolution: Fixed
  • Submitted: 2018-08-15
  • Updated: 2018-08-22
  • Resolved: 2018-08-22
Related Reports
Relates :  
JDK-8209862 - CipherCore performance improvement
Description
There are noticeable performance regressions on both Linux and Windows that need to be investigated. They are each AES related:

��� micro.benchmarks.crypto.small.AESBench.decrypt-algorithm:AES_CBC_NoPadding-dataSize:1024-keyLength:128-provider:
      LXS:   approx. -18%
      WXS:  approx. -3%

��� micro.benchmarks.crypto.small.AESBench.decrypt-algorithm:AES_ECB_NoPadding-dataSize:1024-keyLength:128-provider:
      LXS:   approx. -8%
      WXS:  approx. -4%

��� micro.benchmarks.crypto.small.AESGCMBench.decrypt-algorithm:AES_GCM_NoPadding-dataSize:1024-keyLength:128-provider:
      LXS:   approx. -3%
      WXS:  approx. -2%

The regressions reproduce when ran on the same machine (ie, triage). 
The regressions reproduce when ran on a different machine (ie, re-traige). 
A JVMswap run indicates the cause is in the JDK (ie, not the JVM).
Comments
Source of the regression found. Some improvement was suggested.
22-08-2018
Performance improvement was suggested https://bugs.openjdk.java.net/browse/JDK-8209862
22-08-2018
The source of regression is security fix for https://bugs.openjdk.java.net/browse/JDK-8207775 which required to add zeroing sensitive data operations.
22-08-2018