spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kazuaki Ishizaki (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-12620) Proposal of GPU exploitation for Spark
Date Mon, 04 Jan 2016 16:38:40 GMT

    [ https://issues.apache.org/jira/browse/SPARK-12620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15081332#comment-15081332
] 

Kazuaki Ishizaki commented on SPARK-12620:
------------------------------------------

[~srowen] would it be better to continue discussion on SPARK-3785 while we keep the JIRA entry
as RESOLVED or CLOSED? As [~rxin] said, SPARK-3785 may be used as a venue for discussion.

> Proposal of GPU exploitation for Spark
> --------------------------------------
>
>                 Key: SPARK-12620
>                 URL: https://issues.apache.org/jira/browse/SPARK-12620
>             Project: Spark
>          Issue Type: New Feature
>          Components: Spark Core
>            Reporter: Kazuaki Ishizaki
>
> I created a new JIRA entry to move from SPARK-3875
> Exploiting GPUs can allow us to shorten the execution time of a Spark job and to reduce
the number of machines in a cluster. We are working to effectively and easily exploit GPUs
on Spark at  [http://github.com/kiszk/spark-gpu]. Our project page is [http://kiszk.github.io/spark-gpu/].
A design document is [here|https://docs.google.com/document/d/1bo1hbQ7ikdUA9LYtYh6kU_TwjFK2ebkHsH66QlmbYP8/edit?usp=sharing]
> Our ideas for exploiting GPUs are
> # adding a new format for a partition in an RDD, which is a column-based structure in
an array format, in addition to the current Iterator\[T\] format with Seq\[T\]
> # generating parallelized GPU native code to access data in the new format from a Spark
application program by using an optimizer and code generator (this is similar to [Project
Tungsten|https://databricks.com/blog/2015/04/28/project-tungsten-bringing-spark-closer-to-bare-metal.html])
and pre-compiled library
> The motivation of idea 1 is to reduce the overhead of serializing/deserializing partition
data for copy between CPU and GPU. The motivation of idea 2 is to avoid writing hardware-dependent
code by application programmers. At first, we are working for idea A (For idea B, we need
to write [CUDA|https://en.wikipedia.org/wiki/CUDA] code for now). 
> This prototype achieved [3.15x performance improvement|https://github.com/kiszk/spark-gpu/wiki/Benchmark]
of logistic regression ([SparkGPULR|https://github.com/kiszk/spark-gpu/blob/dev/examples/src/main/scala/org/apache/spark/examples/SparkGPULR.scala])
in examples on a 16-thread IvyBridge box with an NVIDIA K40 GPU card over that with no GPU
card
> You can download the pre-build binary for x86_64 and ppc64le from [here|https://github.com/kiszk/spark-gpu/wiki/Downloads].
You can run this on Amazon EC2 by [the procedure|https://github.com/kiszk/spark-gpu/wiki/How-to-run-%28local-or-AWS-EC2%29],
too.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org


Mime
View raw message