[ https://issues.apache.org/jira/browse/HDFS-1094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12873238#action_12873238 ]
dhruba borthakur commented on HDFS-1094: ---------------------------------------- >From all the above mathematical formulae, can we then say that the policy >listed below can be a good first step: "The first replica is on a node on rack r. Then the other two replicas be randomly selected nodes on either rack r-1 or r+1. In this approach, three datanodes in two consecutive racks have to fail simultaneously for a block loss to occur. This is better than the current implementation where any three datanode failures in the entire cluster can cause one block to be lost." The NetworkTopology class in the namenode already has information about node to rack location. > 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.