[
https://issues.apache.org/jira/browse/IGNITE-610?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14482993#comment-14482993
]
Sergey Evdokimov commented on IGNITE-610:
-----------------------------------------
Result of my investigation:
Offheap index slower then onheap at 30%.
I've created benchmark based on sql-query-onheap benchmark to investigation.
All performance measurements made on it with 1 node and 1 thread.
Benchmark result:
offheap: 11800 op/sec , onheap 15800 op/sec
I've found that main cause of the slowdown is size of objects cache. I set it
unlimited for both onheap and offheap. benchmark result:
offheap: 14200 op/sec , onheap 16200 op/sec
On CPU usage snapshot a lot of time get GridOffHeapSnapTreeMap#close(), or
rather "mem.release(holderRef, NODE_SIZE)". Each query execution takes snapshot
of tree, it allocates and deallocates memory. I've temporary disabled getting
of snapshot, benchmark results:
offheap: 15200 op/sec , onheap 16200 op/sec
The remaining 6% is an overhead for offheap memory access.
> Need to improve offheap query performance
> -----------------------------------------
>
> Key: IGNITE-610
> URL: https://issues.apache.org/jira/browse/IGNITE-610
> Project: Ignite
> Issue Type: Task
> Reporter: Yakov Zhdanov
> Assignee: Sergey Evdokimov
> Priority: Blocker
> Fix For: sprint-3
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
