[
https://issues.apache.org/jira/browse/HBASE-12934?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14294171#comment-14294171
]
Lars Hofhansl commented on HBASE-12934:
---------------------------------------
Yep... Seen that. We also have the bandwidth limit from the write pipeline.
125mb/s max for 1ge. 1.56gb/s for 10ge. Also I think we usually say that an
HDFS output stream can write at about 40-50mb/s. Maybe an HDFS guy can provide
some actual numbers.
So with really nice SSDs and 10ge networking between racks, I can see an
advantage, if HDFS can handle, but only if we do this for flushes and
compactions (i.e. the CF is essentially stored on SSDs).
Then again compactions are tricky, since the write-amplification will wear out
the SSDs.
Bottom line, this should be backed up by some perf data :)
And as I stated, other parts are more obvious: finishing HBASE-5954 and then
place WALs on SSDs, or allow column families to be stored on SSDs for random
read load.
Why don't we do that first?
Happy to meet up on this and have face-2-face discussion.
> Utilize Flash storage for flushing
> ----------------------------------
>
> Key: HBASE-12934
> URL: https://issues.apache.org/jira/browse/HBASE-12934
> Project: HBase
> Issue Type: Sub-task
> Reporter: Ted Yu
> Assignee: Ted Yu
> Attachments: 12934-001.txt
>
>
> Store flushing should be able to make use of hdfs storage policy.
> One option is to allow setting storage policy for the directory path of the
> specified column family.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
