hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Karthik Ranganathan (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HDFS-1094) Intelligent block placement policy to decrease probability of block loss
Date Mon, 12 Apr 2010 18:17:50 GMT

    [ https://issues.apache.org/jira/browse/HDFS-1094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12856097#action_12856097

Karthik Ranganathan commented on HDFS-1094:

The issue I see with 1 (which may not be a big deal) is that if more racks are added, then
the placement policy cannot utilize the new, emptier machines more efficiently without re-balancing
data. Would a hybrid help? Replica 1 is chosen at random from (r-1) and the next replica using
scheme 2 (excluding the node on rack r-1).

I think this change protects against node failures more than rack failures. Lets say the replication
factor is three. If you calculate the probability of data loss if *any* 3 machines fail in
the cluster at the same time, then the placement matters:

1) If the blocks are scattered across all the machines, then any 3 machines failing will contribute
towards data loss. 
2) If a given block (by virtue of the new replication scheme) is placed only on a subset of
machines, then the probability of data loss is reduced from any 3 machines failing to 3 machines
in that subset failing. Failures of machines in the distinct subsets don't contribute to data

> 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: dhruba borthakur
> 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.
If you think it was sent incorrectly contact one of the administrators: https://issues.apache.org/jira/secure/Administrators.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message