[
https://issues.apache.org/jira/browse/HBASE-17021?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15646938#comment-15646938
]
ramkrishna.s.vasudevan commented on HBASE-17021:
------------------------------------------------
I just tried out a PE run without checksum. I set dfs.checksum.type to NULL. I
still see AsyncWAL is twice slower.
bq. But I think the copies when ByteBuf grow maybe expensive as the data we
need to copy will also increase. Maybe we should set a larger initial size
instead of using the default 256B.
I could see some Unsafe copies that are prominent now. Let me try with a patch
for that.
> Use RingBuffer to reduce the contention in AsyncFSWAL
> -----------------------------------------------------
>
> Key: HBASE-17021
> URL: https://issues.apache.org/jira/browse/HBASE-17021
> Project: HBase
> Issue Type: Sub-task
> Components: wal
> Affects Versions: 2.0.0
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Fix For: 2.0.0
>
> Attachments: AsyncWAL_disruptor_7.patch, HBASE-17021-v1.patch,
> HBASE-17021-v2.patch, HBASE-17021.patch
>
>
> The WALPE result in HBASE-16890 shows that with disruptor's RingBuffer we can
> get a better performance.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
