mbien commented on code in PR #4284:
URL: https://github.com/apache/netbeans/pull/4284#discussion_r906847815
##########
nbbuild/templates/projectized.xml:
##########
@@ -290,7 +290,7 @@
<target name="test-lib-init"
depends="-init-bootclasspath-prepend,init,-build-libs.junit4">
<path id="test.unit.lib.cp"/>
<property name="test.jms.flags" value=""/>
- <property name="test.run.args" value="-ea -Xmx700m
${metabuild.jms-flags.jvm} ${test.jms.flags} -XX:+IgnoreUnrecognizedVMOptions"/>
+ <property name="test.run.args" value="-ea -Xms1200m -Xmx1200m
-XX:+UseParallelGC ${metabuild.jms-flags.jvm} ${test.jms.flags}
-XX:+IgnoreUnrecognizedVMOptions"/>
Review Comment:
Hi Matthias,
ParallelGC is a generational throughput GC. It is available on all JDKs from
all hotspot vendors and designed for tasks which are willing to trade latency
for throughput (batch processing or tests). The GC can do that since it is not
concurrent, and has therefore no additional runtime overhead (for both CPU and
heap upkeep overhead).
G1 (new default), Shenandoha and ZGC are the concurrent GCs which try to
keep latency low but have more overhead. G1 is general purpose, ZGC is mostly
for very large heaps, Shenandoah scales well from small to large heaps and has
also lower heap size overhead compared to other concurrent GCs, but is not
included in Oracle JDKs. (picking between the 3 concurrent GCs is often
application specific and requires testing)
SerialGC is like parallel just single threaded, nice for small heaps, e.g
embedded or single threaded micro service deployment (Shenandoah is also nice
for embedded if you have multiple cores and low throughput).
Our nodes (both gh actions and travis have 2 threads minimum, and 7GB RAM
(travis might have 4GB)) so we can use ParallelGC instead of SerialGC. Further,
increasing heap size even allows to run some tests without triggering a GC
cycle at all. By using a non-concurrent GC we also get lower overhead, even if
it doesn't run. We could likely further tweak this by increasing Xmx, but the
gain might be fairly small.
Did I measure that for the tests? No since we have 14h of test jobs. But
since we are looking for throughput and don't care about latency the choice is
fairly simple since there is only one GC available on hotspot distributions
which is made for this usecase.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
For further information about the NetBeans mailing lists, visit:
https://cwiki.apache.org/confluence/display/NETBEANS/Mailing+lists
