[cassandra] OOM under heavy load

After the migration to the G1, the server are now subject to OOM. They are under heavy load (lot of replayers, big repair in progress)

Dec 29 14:39:47 cassandra01 cassandra[777199]: OpenJDK 64-Bit Server VM warning: INFO: os::commit_memory(0x00007fa718b13000, 16384, 0) failed; error='Not enough space' (errno=12)
Dec 29 14:39:47 cassandra01 cassandra[777199]: #
Dec 29 14:39:47 cassandra01 cassandra[777199]: # There is insufficient memory for the Java Runtime Environment to continue.
Dec 29 14:39:47 cassandra01 cassandra[777199]: # Native memory allocation (mmap) failed to map 16384 bytes for committing reserved memory.
Dec 29 14:39:47 cassandra01 cassandra[777199]: # An error report file with more information is saved as:
Dec 29 14:39:47 cassandra01 cassandra[777199]: # /tmp/hs_err_pid777199.log
Dec 29 14:39:47 cassandra01 cassandra[777199]: [thread 828675 also had an error][thread 828704 also had an error][thread 843094 also had an error]
Dec 29 14:39:47 cassandra01 cassandra[777199]: [thread 1033436 also had an error][thread 1007071 also had an error][thread 1087563 also had an error][thread 841719 also had an error]
Dec 29 14:39:47 cassandra01 cassandra[777199]: [thread 1087561 also had an error]
Dec 29 14:40:00 cassandra01 systemd[1]: cassandra@instance1.service: Main process exited, code=exited, status=1/FAILURE
Dec 29 14:40:00 cassandra01 systemd[1]: cassandra@instance1.service: Failed with result 'exit-code'.