[infinispan-issues] [JBoss JIRA] (ISPN-2414) Optimise memory hotspots for local caches

Optimise memory hotspots for local caches ----------------------------------------- Key: ISPN-2414 URL: https://issues.jboss.org/browse/ISPN-2414 Project: Infinispan Issue Type: Enhancement Reporter: Galder Zamarreño Assignee: Galder Zamarreño Fix For: 5.2.0.CR1, 5.2.0.Final, 5.1.9.Final Attachments: Screen Shot 2012-10-17 at 1.21.48 PM.png, Screen Shot 2012-10-17 at 1.25.45 PM.png, Screen Shot 2012-10-17 at 1.30.09 PM.png, Screen Shot 2012-10-17 at 1.33.46 PM.png Several small memory hotspots have been discovered during some performance testing of local caches. Optimise them. 1. Collections.EmptySet.Iterator: Screen%20Shot%202012-10-17%20at%201.21.48%20PM.png - First of all, iterating over an empty set returns a brand new iterator instance all the time. WTF? - Second, why does get() have to call unlockAll()? We do not acquire locks on get (unless it has a FORCE_WRITE_LOCK flag on it), right? 2. ThreadLocal.set: Screen%20Shot%202012-10-17%20at%201.25.45%20PM.png - Is this neeeded at all? - If it's needed for any use case, can it be optimised away to avoid using it for a local cache where no marshaller is used, and where no cluster cache loader is used? 3. AbstractInvocationContext instances: Screen%20Shot%202012-10-17%20at%201.30.09%20PM.png - Why is this expensive? Is it cos the reference to the ClassLoader? - If it's the ClassLoader ref, can we move this away for local caches? 4. AbstractFlagAffectedCommand instances: Screen%20Shot%202012-10-17%20at%201.33.46%20PM.png - Might be worth thinking how can we shrink these commands for local, unflagged, operations, such as get?