[jira] Updated: (MAPREDUCE-64) Map-side sort is hampered by io.sort.record.percent

["Chris Douglas (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/MAPREDUCE-64?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Chris Douglas updated MAPREDUCE-64:
-----------------------------------

    Attachment: M64-0.patch

Preliminary patch, merged with trunk. Still testing, evaluating performance, and verifying correctness.

This modifies the existing sort as follows. The kvoffsets and kvindices arrays are merged and interlaced into the serialization buffer, read/written through an IntBuffer overlay. All collection proceeds from an "equator"; record serialization moves "forward" in the buffer while metadata moves "backward," both as circular buffers as in the existing design. The "io.sort.record.percent" parameter is no longer used. The spill percentage still controls the soft limit, which resets the equator into the unused part of the buffer at an offset proportional to the average record size.

> Map-side sort is hampered by io.sort.record.percent
> ---------------------------------------------------
>
>                 Key: MAPREDUCE-64
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-64
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>            Reporter: Arun C Murthy
>         Attachments: M64-0.patch
>
>
> Currently io.sort.record.percent is a fairly obscure, per-job configurable, expert-level parameter which controls how much accounting space is available for records in the map-side sort buffer (io.sort.mb). Typically values for io.sort.mb (100) and io.sort.record.percent (0.05) imply that we can store ~350,000 records in the buffer before necessitating a sort/combine/spill.
> However for many applications which deal with small records e.g. the world-famous wordcount and it's family this implies we can only use 5-10% of io.sort.mb i.e. (5-10M) before we spill inspite of having _much_ more memory available in the sort-buffer. The word-count for e.g. results in ~12 spills (given hdfs block size of 64M). The presence of a combiner exacerbates the problem by piling serialization/deserialization of records too...
> Sure, jobs can configure io.sort.record.percent, but it's tedious and obscure; we really can do better by getting the framework to automagically pick it by using all available memory (upto io.sort.mb) for either the data or accounting.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.