[
https://issues.apache.org/jira/browse/IGNITE-2968?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15415170#comment-15415170
]
Andrey Gura commented on IGNITE-2968:
-------------------------------------
Transactional benchmarks show performance drop about 2-5%. Change that leads to
this performance drop is replacing of {{LinkedHashMap}} by
{{ConcurrentLinkedHashMap}}. I think it is related with with concurrent map
allocation because most benchmarks envolve only one entry in tansaction.
Nevertheless, I've replaced concurrent map by non-concurrent map again, made
{{IgniteTxStateImpl.addEntry()}} method synchronized and return entries copy
form tx state in order to avoid {{ConcurrentModificationException}}. The lock
introduced by synchronized method should be thin because optimistic
transactions enlist entries in one thread before any request will be send and
pessimistic transactions enlist entries on user operation also before sending
any requests. So this lock should not have any contention unless deadlock
detection procedure started.
Waiting for benchmarks results.
> Near cache support in transactions deadlock detection
> -----------------------------------------------------
>
> Key: IGNITE-2968
> URL: https://issues.apache.org/jira/browse/IGNITE-2968
> Project: Ignite
> Issue Type: Improvement
> Components: cache
> Reporter: Andrey Gura
> Assignee: Andrey Gura
> Fix For: 1.8
>
>
> Deadlock detection doesn't support transactions on near cache. Need implement
> it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
