[
https://issues.apache.org/jira/browse/HADOOP-3524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12603871#action_12603871
]
Brice Arnould commented on HADOOP-3524:
---------------------------------------
> I ran Brice's test and following are the results
OK, I must have misunderstood something.
I'm going to explain my understing of this subject, but it's likely that I
missed something. Anyway, we'll see ^_^
A first thing is that my tests do not compare the time of both approaches, but
the time needed to call currentTimeMillis to the time needed to access a field.
And it was more "for fun", because my impressions is that currentTimeMillis()
already takes a negligible time. Even if we could make it to take 0
picoseconds, we would spare about 14ms every 10000 calls. That, is arround 14ms
every 1 minute and 16 seconds, at the extreme rate of 130 request by second.
And that is the extreme situation.
There is a need to make things more parallel in the JobTracker, but in my
opinion currentTimeMillis() do not worth it.
That being said, I probably missed something :-P
> JobTracker's processHeartbeat() should not call System.currentTimeMillis()
> everytime
> ------------------------------------------------------------------------------------
>
> Key: HADOOP-3524
> URL: https://issues.apache.org/jira/browse/HADOOP-3524
> Project: Hadoop Core
> Issue Type: Improvement
> Components: mapred
> Reporter: Amar Kamat
> Attachments: CurrentTimeCost.java
>
>
> Consider the following
> {code:title=JobTracker.java|borderStyle=solid}
> private synchronized boolean processHeartbeat(
> TaskTrackerStatus
> trackerStatus, boolean initialContact) {
> String trackerName = trackerStatus.getTrackerName();
> trackerStatus.setLastSeen(System.currentTimeMillis());
> {code}
> Here, the call to {{System.currentTimeMillis()}} on every call to
> {{JobTracker.processHeartbeat()}} might prove costly. While
> testing/benchmarking HADOOP-2119, we recorded that the JobTracker was able to
> serve ~130 tasks/sec. So that means we might make ~130 calls to
> {{System.currentTimeMillis()}} per second. I think in these cases
> (_last-seen-status_ etc) such a high level of accuracy in terms of timestamp
> is unnecessary and hence can be avoided.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
