hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tsz Wo Nicholas Sze (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-12916) Allow different Hadoop IPC Call Queue throttling policies with FCQ/BackOff
Date Wed, 30 Mar 2016 13:08:25 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-12916?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15217939#comment-15217939
] 

Tsz Wo Nicholas Sze commented on HADOOP-12916:
----------------------------------------------

{code}
+        double decayedAvgRespTime = (responseTimeAvgInLastWindow.get(i) > 0.0) ?
+            decayFactor * responseTimeAvgInLastWindow.get(i) +
+            (1 - decayFactor) * averageResponseTime : averageResponseTime;
{code}
For the decayed case, should the formula be {{decayFactor*responseTimeAvgInLastWindow.get(
i) + averageResponseTime}}, i.e. no {{(1 - decayFactor)}} in the second term?

BTW, the if-statement can be rewritten as below to make it shorter.
{code}
      final double lastAvg = responseTimeAvgInLastWindow.get(i);
      if (enableDecay && lastAvg > 0) {
        final double decayed =  decayFactor * lastAvg + averageResponseTime;
        responseTimeAvgInLastWindow.set(i, decayed);
      } else {
        responseTimeAvgInLastWindow.set(i, averageResponseTime);
      }
{code}


> Allow different Hadoop IPC Call Queue throttling policies with FCQ/BackOff
> --------------------------------------------------------------------------
>
>                 Key: HADOOP-12916
>                 URL: https://issues.apache.org/jira/browse/HADOOP-12916
>             Project: Hadoop Common
>          Issue Type: Improvement
>          Components: ipc
>            Reporter: Xiaoyu Yao
>            Assignee: Xiaoyu Yao
>         Attachments: HADOOP-12916.00.patch, HADOOP-12916.01.patch, HADOOP-12916.02.patch,
HADOOP-12916.03.patch, HADOOP-12916.04.patch, HADOOP-12916.05.patch
>
>
> Currently back off policy from HADOOP-10597 is hard coded to base on whether call queue
is full. This ticket is open to allow flexible back off policies such as moving average of
response time in RPC calls of different priorities. 



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

Mime
View raw message