[
https://issues.apache.org/jira/browse/HDFS-1094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12878534#action_12878534
]
Rodrigo Schmidt commented on HDFS-1094:
---------------------------------------
For this JIRA I'm implementing a general idea, in which configuration variables
determine a subset of the racks that can be used for each block and a subset of
machines within that rack.
However, the NetworkTopology class from Hadoop Common lacks methods to list
racks or limit machines within a rack. I think this is fundamental for this
JIRA or any other complex block placement policy (NetworkTopology has a very
limited API).
Even though it's possible to circumvent the NetworkTopology limitations by
accessing the namenode directly (through FSNameSystem passed as FSClusterStats
to BlockPlacementPolicy), this makes the code really ugly.
I would like to get the general opinion about extending NetworkTopology from
hadoop common.
> Intelligent block placement policy to decrease probability of block loss
> ------------------------------------------------------------------------
>
> Key: HDFS-1094
> URL: https://issues.apache.org/jira/browse/HDFS-1094
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: name-node
> Reporter: dhruba borthakur
> Assignee: Rodrigo Schmidt
> Attachments: prob.pdf, prob.pdf
>
>
> The current HDFS implementation specifies that the first replica is local and
> the other two replicas are on any two random nodes on a random remote rack.
> This means that if any three datanodes die together, then there is a
> non-trivial probability of losing at least one block in the cluster. This
> JIRA is to discuss if there is a better algorithm that can lower probability
> of losing a block.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
