[ https://issues.apache.org/jira/browse/HDFS-15452?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17154314#comment-17154314 ]
Xiaoqiao He commented on HDFS-15452: ------------------------------------ Thanks involve me here. Sorry I don't get what issues do you meet. IIUC, tuning the percentage of memory capacity will impact performance obviously (GSet collision ratio or heap occupy). I don't trace why this is 2% rather than other percentage since it is very historical logic. About patch [^HDFS-15452.001.patch], a. it is better to give value section, rather than any value will pass now. b. does `INodeMap`, `RetryCache` and `cachedBlocks` also need to config? I prefer to keep this static value, in my practice it runs well for over 600GB heap memory. I don't find any heap occupy cost or performance issue. > Dynamically initialize the capacity of BlocksMap > ------------------------------------------------ > > Key: HDFS-15452 > URL: https://issues.apache.org/jira/browse/HDFS-15452 > Project: Hadoop HDFS > Issue Type: Improvement > Affects Versions: 3.0.3 > Reporter: jianghua zhu > Assignee: jianghua zhu > Priority: Major > Attachments: HDFS-15452.001.patch > > > The default value for initializing BlocksMap in the BlockManager class is 2. > This can be set to a dynamic value. > BlockManager#BlockManager() { > ...... > // Compute the map capacity by allocating 2% of total memory > blocksMap = new BlocksMap( > LightWeightGSet.computeCapacity(2.0, "BlocksMap")); > ...... > } > -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org