[ https://issues.apache.org/jira/browse
/HADOOP-1900?page=com.atlassian.jira.plugin.system.issuetabp
anels:comment-tabpanel#action_12535509 ]
Amareshwari Sri Ramadasu commented on HADOOP-1900:
--------------------------------------------------
Here is a proposal for changing heartbeat interval
dynamically.
We will intialize heartbeat interval as
HEARTBEAT_INTERVAL(10 secs) in task tracker.
Once tasktracker transmits heartbeat, jobtracker's response
will have next heartbeat interval.
JobTracker calculates next heart beat interval as follows:
1. Using Clustersize (Number of task trackers) :
nextInterval = 10secs * (cluster_size/500 +1)
i.e. For every additional 500 nodes we increase
heartbeat interval by 10 secs.
2. if (nextInterval >HEARTBEAT_INTERVAL_MAX ) nextInteval
= HEARTBEAT_INTERVAL_MAX;
if this exceeds HEARTBEAT_INTERVAL_MAX (60seconds), next
heartbeat interval is 60 seconds.
3. if dropcount of heartbeats is greater than a threshold
drop count, increase interval by 10 more seconds.
Threshold drop count can be 'clustersize/10' . i.e. If
there are 500 nodes in the cluster and more than 50
heartbeats are dropped, then we increase next heartbeat
interval by 10 more seconds.
Thus, next Interval calculation can be the following
threshold_dropcount = clustersize/10;
isBusy = dropCount > threshold_dropcount ?1:0;
nextInterval = HEARTBEAT_INTERVAL* (cluster_size/500 +1)
+
HEARTBEAT_INTERVAL*isBusy;
if (nextInterval >HEARTBEAT_INTERVAL_MAX ) nextInteval =
HEARTBEAT_INTERVAL_MAX;
Now, MapEventsFetcherThread polls jobtracker for completed
map tasks for every 5 secs (MIN_POLL_INTERVAL). Shall we
change polling interval also in the similar fashion as
heartbeat interval? But, here some reduce tasks could be
idle for longer time.
Any thoughts?
> the heartbeat and task event queries interval should be
set dynamically by the JobTracker
>
------------------------------------------------------------
-----------------------------
>
> Key: HADOOP-1900
> URL: htt
ps://issues.apache.org/jira/browse/HADOOP-1900
> Project: Hadoop
> Issue Type: Improvement
> Components: mapred
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
>
> The JobTracker should scale the intervals that the
TaskTrackers use to contact it dynamically, based on how the
busy it is and the size of the cluster.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue
online.
|
