Mailing-List: contact issues-help@flink.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@flink.incubator.apache.org
Date: Sat, 1 Nov 2014 19:35:33 +0000 (UTC)
From: "Stephan Ewen (JIRA)" <jira@apache.org>
To: issues@flink.incubator.apache.org
Message-ID: <JIRA.12751234.1414529931000.391540.1414870533659@Atlassian.JIRA>
In-Reply-To: <JIRA.12751234.1414529931000@Atlassian.JIRA>
References: <JIRA.12751234.1414529931000@Atlassian.JIRA>
 <JIRA.12751234.1414529931166@arcas>
Subject: [jira] [Commented] (FLINK-1198) Store broadcast variables once per
 TaskManager
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/FLINK-1198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14193391#comment-14193391 ] 

Stephan Ewen commented on FLINK-1198:
-------------------------------------

I have worked in this issue already for the past week and have it running (including iterations) except for compatibility with the deprecated record API. 

I think it easier if I just finish that part, since the tricky cases are all done.

> Store broadcast variables once per TaskManager
> ----------------------------------------------
>
>                 Key: FLINK-1198
>                 URL: https://issues.apache.org/jira/browse/FLINK-1198
>             Project: Flink
>          Issue Type: Improvement
>          Components: Local Runtime
>    Affects Versions: 0.7.0-incubating
>            Reporter: Stephan Ewen
>            Assignee: Alexander Alexandrov
>             Fix For: 0.8-incubating
>
>
> Currently, a copy of the broadcast variable is sent to each parallel task, rather than once to the TaskManager where the tasks share them.
> Sharing the braodcast variables would increase memory efficiency significantly.


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