[ https://issues.apache.org/jira/browse/HDFS-8876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14662306#comment-14662306 ]
Tsz Wo Nicholas Sze commented on HDFS-8876: ------------------------------------------- Thanks for filing the JIRAs. I already have HDFS-8818 to speed up Balancer. In my tests, it can speed up from 4GB per iteration to 800GB per iteration while the duration of each iteration is only increased from ~1 minutes to 1-2 minutes. I also have patches to remove some useless hard coded parameters and make the other hard coded parameters configurable; see HDFS-8818 and HDFS-8824. > Make hard coded parameters used by balancer and other tools configurable > ------------------------------------------------------------------------ > > Key: HDFS-8876 > URL: https://issues.apache.org/jira/browse/HDFS-8876 > Project: Hadoop HDFS > Issue Type: Sub-task > Reporter: Ming Ma > Assignee: Chris Trezzo > > During investigation of how to speed up balancer, at least to the level > specified by {{dfs.datanode.balance.bandwidthPerSec}}, we found that > parameters such as {{MAX_BLOCKS_SIZE_TO_FETCH}} and > {{SOURCE_BLOCKS_MIN_SIZE}} are hard coded. These parameters are related to > block size and other configurable parameters used by balancer. So least we > should make it configurable. In the longer term, it might be interesting to > understand if we simplify all these related configurations. -- This message was sent by Atlassian JIRA (v6.3.4#6332)