You want to decrease your young gen (defaults to 40% of heap, which is *way* to big for HBase). I wrote the reasoning here: http://hadoop-hbase.blogspot.com/2014/03/hbase-gc-tuning-observations.html
(Basically HBase produces a lot of day-to-day garbage that can be collected quickly. You do not want to accumulate a GC debt for that kind of garbage and hence keep the young gen *very* small, like 256-512mb or so) Let us know if that fixes your problem? -- Lars ________________________________ From: yanivG <[email protected]> To: [email protected] Sent: Tuesday, August 19, 2014 10:13 PM Subject: Re: GC peaks during major compaction Yes. The RegionServer cmd arguments are: /usr/java/latest//bin/java -XX:OnOutOfMemoryError=kill -9 %p -Xmx30720m -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+UseNUMA -XX:CMSInitiatingOccupancyFraction=60 -ea -XX:+HeapDumpOnOutOfMemoryError -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+UseNUMA -XX:CMSInitiatingOccupancyFraction=60 -ea -XX:+HeapDumpOnOutOfMemoryError free -g output: total used free shared buffers cached Mem: 126 114 11 0 0 90 -/+ buffers/cache: 23 102 Swap: 9 0 9 meminfo MemTotal: 132145596 kB MemFree: 11875380 kB Buffers: 359804 kB Cached: 94909280 kB SwapCached: 0 kB Active: 69301468 kB Inactive: 47497184 kB Active(anon): 18676840 kB Inactive(anon): 2858304 kB Active(file): 50624628 kB Inactive(file): 44638880 kB -- View this message in context: http://apache-hbase.679495.n3.nabble.com/GC-peaks-during-major-compaction-tp4062704p4062930.html Sent from the HBase User mailing list archive at Nabble.com.
