[jboss-jira] [JBoss JIRA] (JGRP-1396) NAKACK2: merge NakReceiverWindow and Retransmitter

[ https://issues.jboss.org/browse/JGRP-1396?page=com.atlassian.jira.plugin.... ] Bela Ban commented on JGRP-1396: -------------------------------- OK, so I came up with this lockless RingBuffer, but then I created a copy, this time using a lock to synchronize access to add(), removeMany() etc, and the synchronized version was faster than the lockless version ! I guess, if I compare the 2: lockless: --------- public void add() { // read volatile HD // ... // write volatile HR // write volatile HD } synchronized: ------------- public void add() { synchronized { // read HD (non-volatile) // ... // write HR // write HD } } The lockless version has lots of reads and writes to the same volatile fields (memory barriers), so most of the reads will not find their value in the cache and thus have to read the value from memory (or a 2nd level cache). Most writes will flush the cache, so this is slow. Plus, if we have false sharing (e.g. HD and LOW on the same cache line), then this gets compounded... OTOH, the synchronized version does a read before the synchronization and a flush when exiting it. The cost of acquiring a lock is apparently smaller than the many memory barriers that get executed ! -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.jboss.org/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira