[jira] [Commented] (HBASE-16717) Reuse DataChecksum
[ https://issues.apache.org/jira/browse/HBASE-16717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15525971#comment-15525971 ] binlijin commented on HBASE-16717: -- Do with a minor test, it is hard too say which is the best, so close this. > Reuse DataChecksum > -- > > Key: HBASE-16717 > URL: https://issues.apache.org/jira/browse/HBASE-16717 > Project: HBase > Issue Type: Improvement >Reporter: binlijin >Priority: Minor > Attachments: HBASE-16717-master.patch > > > Every time to generateChecksums or validateChecksum will create a new > DataChecksum, and DataChecksum is not thread safe, so use ThreadLocal to > reuse it and keep thread safe. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-16717) Reuse DataChecksum
[ https://issues.apache.org/jira/browse/HBASE-16717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15525386#comment-15525386 ] Anoop Sam John commented on HBASE-16717: So we avoid 2 objects creation (The Checksum object per type and DataCheckSum object). But we do have a ThreadLocal read. The ThreadLocal comes with overhead. So am not sure abt the gain of 2 object not getting created is worth enough. I feel it is ok to create object as per demand only than going with TL. Ya the allocation u wil see some gain.. But on net perf I dont think so. The GC should be able to clean these objects fast. Pls check > Reuse DataChecksum > -- > > Key: HBASE-16717 > URL: https://issues.apache.org/jira/browse/HBASE-16717 > Project: HBase > Issue Type: Improvement >Reporter: binlijin >Priority: Minor > Attachments: HBASE-16717-master.patch > > > Every time to generateChecksums or validateChecksum will create a new > DataChecksum, and DataChecksum is not thread safe, so use ThreadLocal to > reuse it and keep thread safe. -- This message was sent by Atlassian JIRA (v6.3.4#6332)