如何减少Java GC年轻大小的暂停时间

Question

I am trying to write Java program to understand gc optimization. I am newbie in java. The program read huge partitons key from cassandra concurrently, let's say 2 parallelism. After reading, i have nothing to do with the rows. They will be waiting GC. After reading 6 partition key, First GC runs and takes 10 seconds. In my opinion, first 4 partition key are being removed.

My problem is, 10 seconds gc pause two much, i use GC setting like below. Is there any suggestion?

java -Dcom.sun.management.jmxremote \
  -Dcom.sun.management.jmxremote.port=9010 \
  -Dcom.sun.management.jmxremote.local.only=false \
  -Dcom.sun.management.jmxremote.authenticate=false \
  -Dcom.sun.management.jmxremote.ssl=false \
  -Xms24g \
  -Xmx24g \
  -XX:+PrintGC \
  -XX:+UseG1GC \
  -XX:MaxGCPauseMillis=200 \
  -XX:G1HeapRegionSize=6 \
  -XX:ParallelGCThreads=12 \
  -XX:ConcGCThreads=4 \
  -XX:+UnlockExperimentalVMOptions \
  -XX:G1NewSizePercent=70 \
  -XX:G1MaxNewSizePercent=80 \
  -XX:InitiatingHeapOccupancyPercent=30 \
  -jar mapreducer-1.0-SNAPSHOT.jar

Answer 1

In theory, the way to reduce GC pause times for young-space collections is to reduce the young / new space size.

In fact, these options:

  -XX:G1NewSizePercent=70 \
  -XX:G1MaxNewSizePercent=80 \

are forcing the young space to be between 70% and 80% of the total heap size. That is probably why the GC is unable to meet the goal specified by this:

  -XX:MaxGCPauseMillis=200 \

The young space need to be a lot smaller.