RFR: 8180421: Change default value of BiasedLockingStartupDelay to 0
robbin.ehn at oracle.com
Tue Jun 27 14:43:26 UTC 2017
On 06/27/2017 04:10 PM, Claes Redestad wrote:
> Hi Robbin,
> looks good to me!
> While there are some near-regressions in a few startup tests, they are not statistically significant. At the same time the peak performance and latency benefits on some
> benchmarks and applications of not delaying is clearly measurable, which makes this seem like the right thing to do to me.
Thanks and thanks for looking at the numbers!
> On 06/26/2017 03:17 PM, Robbin Ehn wrote:
>> Hi all, please review.
>> On behalf of Stefan J, this patch changes the default value of BiasedLockingStartupDelay to 0.
>> "The delay is however a problem for the some of the GC algorithms that use the mark-word."
>> "Benchmark runs doesn't show any regressions for either startup times or steady state performance when setting it to 0."
>> CSR: https://bugs.openjdk.java.net/browse/JDK-8181778
>> Issue: https://bugs.openjdk.java.net/browse/JDK-8180421
>> diff -r 26a2358e2796 src/share/vm/runtime/globals.hpp
>> --- a/src/share/vm/runtime/globals.hpp Fri Jun 23 15:16:23 2017 -0700
>> +++ b/src/share/vm/runtime/globals.hpp Mon Jun 26 15:03:34 2017 +0200
>> @@ -1307,1 +1307,1 @@
>> - product(intx, BiasedLockingStartupDelay, 4000, \
>> + product(intx, BiasedLockingStartupDelay, 0, \
More information about the hotspot-dev