fapifta commented on a change in pull request #710: HDDS-3173. Provide better default JVM options URL: https://github.com/apache/hadoop-ozone/pull/710#discussion_r398210290
########## File path: hadoop-ozone/dist/src/shell/hdds/hadoop-functions.sh ########## @@ -1541,6 +1541,21 @@ function hadoop_translate_cygwin_path fi } +## @description Adds default GC parameters +## @description Only for server components and only if no other -XX parameters +## @description are set +## @audience private +## @stability evolving +## @replaceable yes +function hadoop_add_default_gc_opts +{ + if [[ "${HADOOP_SUBCMD_SUPPORTDAEMONIZATION}" == true ]]; then + hadoop_debug "Appending default GC parameter to the HADOOP_OPTS" + if [[ ! "$HADOOP_OPTS" =~ "-XX" ]] ; then + HADOOP_OPTS="${HADOOP_OPTS} -XX:ParallelGCThreads=8 -XX:+UseConcMarkSweepGC -XX:CMSInitiatingOccupancyFraction=70 -XX:+CMSParallelRemarkEnabled" Review comment: Which is better heavily depends on the workload, but as the main workload are RPC calls with mostly short lived objects in our server processes, CMS is better due to the easier newGen eviction and low number of tenuring/surviving objects. At least this is the experience from HDFS and our workload on the DN and OM/SCM is very similar to HDFS server roles and consist mainly RPC calls while handling large metadata structures. G1 can be tuned for this kind of workload, but it is not the best performing one for this due to the segmented heap structure. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org