[
https://issues.apache.org/jira/browse/SPARK-9018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14628604#comment-14628604
]
Xiangrui Meng commented on SPARK-9018:
--------------------------------------
The goal is to make it easier to obtain numbers like how much time we spent on
solving least squares in ALS, sorting partitions, and shuffling data. For those
cases, a distributed stopwatch would be quite useful. I'm a little bit worried
about performance if we put distributed stopwatch that uses accumulators at
low-level computations. But if we are careful about it, it should be okay.
> Implement a generic Timer utility for ML algorithms
> ---------------------------------------------------
>
> Key: SPARK-9018
> URL: https://issues.apache.org/jira/browse/SPARK-9018
> Project: Spark
> Issue Type: Sub-task
> Components: ML, MLlib
> Reporter: Xiangrui Meng
> Assignee: Xiangrui Meng
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> The Timer utility should be based on the one implemented in trees. In
> particular, we should offer two versions:
> 1. a global timer that is initialized on the driver and use accumulator to
> aggregate time
> 2. a local timer that is initialized on the worker, and only provide per task
> measurement.
> 1) needs some performance benchmark and guidance on the granularity.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
