Mailing-List: contact issues-help@spark.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@spark.apache.org
Date: Sun, 27 Jul 2014 19:29:38 +0000 (UTC)
From: "Patrick Wendell (JIRA)" <jira@apache.org>
To: issues@spark.apache.org
Message-ID: <JIRA.12729212.1406136106319.52741.1406489378359@arcas>
In-Reply-To: <JIRA.12729212.1406136106319@arcas>
References: <JIRA.12729212.1406136106319@arcas>
Subject: [jira] [Commented] (SPARK-2648) Randomize order of executors when
 fetching shuffle blocks
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/SPARK-2648?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14075719#comment-14075719 ] 

Patrick Wendell commented on SPARK-2648:
----------------------------------------

I was looking more at the current code. From what I can tell we already randomize the order when we are making fetch requests:

https://github.com/apache/spark/blob/9564f8548917f563930d5e87911a304bf206d26e/core/src/main/scala/org/apache/spark/storage/BlockFetcherIterator.scala#L220

> Randomize order of executors when fetching shuffle blocks
> ---------------------------------------------------------
>
>                 Key: SPARK-2648
>                 URL: https://issues.apache.org/jira/browse/SPARK-2648
>             Project: Spark
>          Issue Type: Improvement
>            Reporter: Lianhui Wang
>            Assignee: Lianhui Wang
>            Priority: Critical
>
> like mapreduce we need to shuffle blocksByAddress.it can avoid many reducers to connect a executor at a time.when a map has many paritions, at a time there has so much reduces connecting to this map.so it maybe make network's connect to timeout.
> i created PR for this issue:https://github.com/apache/spark/pull/1549


--
This message was sent by Atlassian JIRA
(v6.2#6252)