[
https://issues.apache.org/jira/browse/IGNITE-2583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15166951#comment-15166951
]
Ilya Lantukh commented on IGNITE-2583:
--------------------------------------
These maps get leaked outside of GridDhtTxLocalAdapter to
GridDhtTxPrepareFuture and GridDhtTxFinishFuture. I think we should keep maps
within single class and add delegating methods to it's interface - it will
simplify all modifications.
On the next step we should create another implementation of
GridDhtTxLocalAdapter for cases when we know beforehand that there will be only
one node mapping entry.
> Investigate CHM8 allocations in GridDhtTxLocalAdapter.
> ------------------------------------------------------
>
> Key: IGNITE-2583
> URL: https://issues.apache.org/jira/browse/IGNITE-2583
> Project: Ignite
> Issue Type: Sub-task
> Components: cache
> Affects Versions: 1.5.0.final
> Reporter: Vladimir Ozerov
> Assignee: Ilya Lantukh
> Fix For: 1.6
>
>
> *Problem*
> Whenever GridDhtTxLocalAdapter is allocated, two CHM8-s are created
> immediately: "nearMap" and "dhtMap".
> This is visible as a memory hotspot, which is responsible for >2% of overall
> allocations during single PUT.
> *Proposed solution*
> 1) Investigate whether we really need concurrent semantics here. Can these
> maps be replaced with HashMap-s or (HashMap + synchronzied)?
> 2) Investigate whether we can optimize for single-mapping scenario.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
