You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by "Greg Hogan (JIRA)" <ji...@apache.org> on 2015/09/10 16:14:45 UTC

[jira] [Created] (FLINK-2655) Minimize intermediate merging of spilled buffers

Greg Hogan created FLINK-2655:
---------------------------------

             Summary: Minimize intermediate merging of spilled buffers
                 Key: FLINK-2655
                 URL: https://issues.apache.org/jira/browse/FLINK-2655
             Project: Flink
          Issue Type: Improvement
          Components: Distributed Runtime
    Affects Versions: master
            Reporter: Greg Hogan


If the number of spilled buffers exceeds taskmanager.runtime.max-fan then the number of files must reduced with an intermediate merge by reading, merging, and spilling into a single, larger file.

The current implementation performs an intermediate merge on all files. An optimal implementation minimizes the amount of merged data by performing partial merges first.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)