JMH 1.5.2

Aleksey Shipilev aleksey.shipilev at oracle.com
Fri Feb 6 11:23:10 UTC 2015


Hi,

JMH 1.5.2 is released and available at Maven Central (thanks to Evgeny,
as usual). This is a minor patch release, and it includes the following
improvements and bugfixes:

 * SingleShot mode had printed the naked timeUnit as it's unit (e.g.
"ms"). That's confusing, and not in line with other results. Changed to
print ".../op":
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901285

 * In some Locales and run configurations, the locale guess was wrong,
and was able to abort the run with UnsupportedEncodingException. The
code was hardened up:
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901287

 * BinaryLinkClient is using a single Java Serialization stream and thus
can accumulate garbage over time. This affects the precise runs with
large amount of iterations, or runs that send lots of messages to the
host VM -- the garbage build up may upset the GC ergonomics, as well as
finally result in OOME. Fixed:
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901288

 * A relevant memory problem is that forked VM was responsible for
accumulating and aggregating the benchmark results. With lots of
iterations, it adds up to a significant heap occupancy. Fixed by sending
the raw iteration results to host VM:
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901290

 * After fixing two previous issues, forked VM is now able to run with
as low as 1-4 Mb per JVM instance. Some tests are added to check we
would not leak in future (pessimistically, the bar is at 16 Mb):
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901291

 * Binary link itself has more buffers now, which improves the
host-forked VM interface performance (really, compensating for increased
traffic from offloading the results storage to the host VM):
  https://bugs.openjdk.java.net/browse/CODETOOLS-7901289

Enjoy!

Thanks,
-Aleksey.



More information about the jmh-dev mailing list