liaoyuxiangqin created HDFS-13883:
-------------------------------------

             Summary: Reduce memory consumption and GC of directory scan
                 Key: HDFS-13883
                 URL: https://issues.apache.org/jira/browse/HDFS-13883
             Project: Hadoop HDFS
          Issue Type: Improvement
    Affects Versions: 3.2.0
            Reporter: liaoyuxiangqin


    When DirectoryScan task have trigger in periodic,  the scan thread to scan 
all disk in this 

DataNode for all blockpool, and construct a ScanInfo per block. So DataNode 
need huge memory to hold those ScanInfo's memory structure when tens of 
millions blocks store in this DataNode.

    Another problem is DataNode implements by java, so DataNode run as a JVM, 
so we need to set a big number for -Xmx to satisfy the memory needs of 
DirectoryScan. But we know the default period of DirectoryScan is 6 hours, and 
at other time DataNode actually need less memory, and JVM can't auto return 
free memory to OS, so many memory utilization rate is low. 

    At last, we have test  close or open DirectoryScan in thirty-millions 
blocks store in the DataNode, and the  -Xmx set is 16G and 32G respectively.So 
i think we can improve the 

DirectoryScan process and save memory such as scan one block pool per period, 
thanks.

   



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org

Reply via email to