[
https://issues.apache.org/jira/browse/TEZ-3202?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15229111#comment-15229111
]
Jonathan Eagles commented on TEZ-3202:
--------------------------------------
The approach taken in the above patch is to create a bridge between the upper
layers and lower layers to reduce the amount of storage needed per segment. I
have seen several cases where the number of segments is near 1000000 and
accounts for almost 300MB in the task attempt heap.
> Reduce the memory need for jobs with high number of segments
> ------------------------------------------------------------
>
> Key: TEZ-3202
> URL: https://issues.apache.org/jira/browse/TEZ-3202
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Jonathan Eagles
> Assignee: Jonathan Eagles
> Attachments: HADOOPPF-10832.2.patch
>
>
> Segment has a 'key' member that holds accounting information to the reader's
> current key buffer, position, and length. There is a 384 byte overhead per
> segment since the account is done with the DataInputBuffer class which
> derives from DataInputStream which has underlying byte[80] and char[80] among
> significant pieces. This jira aims to reduce the overhead per segment
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
