[
https://issues.apache.org/jira/browse/HDFS-15288?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17111271#comment-17111271
]
Ayush Saxena commented on HDFS-15288:
-------------------------------------
Thanx [~elgoiri] for the review.
bq. If we are converting the [0,1] to 0 to 100 in the code, why not just pass 0
to 100 which is usually more intuitive?
I did that to keep the semantics same as {{AvailableSpaceBlockPlacementPolicy}}
so that any person familiar with it can use the same configuration here. Just
to keep the sync. This value is the balance preference, the more it is the more
is the chances that the lesser occupied DN will get chosen, with 0 and 1 being
the extremes.
I open to change this, From Fraction_KEY to PERCENTAGE_KEY, But how to get this
configuration name shortened is quite tough. Let me know, if you have any
suggestion.
Will address other comments, once confirmed about the above one.
> Add Available Space Rack Fault Tolerant BPP
> -------------------------------------------
>
> Key: HDFS-15288
> URL: https://issues.apache.org/jira/browse/HDFS-15288
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Ayush Saxena
> Assignee: Ayush Saxena
> Priority: Major
> Attachments: HDFS-15288-01.patch, HDFS-15288-02.patch
>
>
> The Present {{AvailableSpaceBlockPlacementPolicy}} extends the Default Block
> Placement policy, which makes it apt for Replicated files. But not very
> efficient for EC files, which by default use.
> {{BlockPlacementPolicyRackFaultTolerant}}. So propose a to add new BPP having
> similar optimization as ASBPP where as keeping the spread of Blocks to max
> racks, i.e as RackFaultTolerantBPP.
> This could extend {{BlockPlacementPolicyRackFaultTolerant}}, rather than the
> {{BlockPlacementPOlicyDefault}} like ASBPP and keep other logics of
> optimization same as ASBPP
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
