[
https://issues.apache.org/jira/browse/SYSTEMML-1774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16091874#comment-16091874
]
Mike Dusenberry edited comment on SYSTEMML-1774 at 7/18/17 5:48 PM:
--------------------------------------------------------------------
Also, of course there *shouldn't* be any need to force the REMOTE_SPARK parfor
mode, but that is exactly the reason for this JIRA issue -- for distributed SGD
cases, a plain parfor isn't performing adequately, and this can serve as a
scenario to explore improvements. Instead of executing the parfor op in a
distributed manner across Spark, a plain parfor is running on the driver, and
thus forcing multithreaded convolution ops to run in single threaded mode,
which is counterproductive. [~Tenma] can comment more, but so far he found
that a HYBRID_SPARK + plain parfor setup takes ~1.5 hours, while a SPARK +
plain parfor takes ~30 mins. HYBRID_SPARK + REMOTE_SPARK parfor currently
fails, but I imagine it should be faster than 30 mins.
was (Author: mwdus...@us.ibm.com):
Also, of course there *shouldn't* be any need to force the REMOTE_SPARK parfor
mode, but that is exactly the reason for this JIRA issue -- for distributed SGD
cases, a plain parfor isn't performing adequately. Instead of executing the
parfor op in a distributed manner across Spark, a plain parfor is running on
the driver, and thus forcing multithreaded convolution ops to run in single
threaded mode, which is counterproductive. [~Tenma] can comment more, but so
far he found that a HYBRID_SPARK + plain parfor setup takes ~1.5 hours, while a
SPARK + plain parfor takes ~30 mins. HYBRID_SPARK + REMOTE_SPARK parfor
currently fails, but I imagine it should be faster than 30 mins.
> Improve Parfor parallelism for deep learning
> --------------------------------------------
>
> Key: SYSTEMML-1774
> URL: https://issues.apache.org/jira/browse/SYSTEMML-1774
> Project: SystemML
> Issue Type: Improvement
> Components: Algorithms, Compiler, ParFor
> Affects Versions: SystemML 1.0
> Reporter: Fei Hu
> Labels: deeplearning
> Attachments: Explain_For_HYBRID_SPARK_Mode_With_ErrorInfo.txt,
> Explain_For_Spark_Mode.txt, MNIST_Distrib_Sgd.scala,
> mnist_lenet_distrib_sgd.dml
>
>
> When running the [distributed MNIST LeNet example |
> https://github.com/apache/systemml/blob/master/scripts/nn/examples/mnist_lenet_distrib_sgd.dml],
> each mini-batch could ideally run in parallel without interaction. We try to
> force {{parfor (j in 1:parallel_batches)}} at line 137 of
> {{nn/examples/mnist_lenet_distrib_sgd.dml}} to be {{parfor (j in
> 1:parallel_batches, mode=REMOTE_SPARK, opt=CONSTRAINED)}} use
> {{REMOTE_SPARK}} mode, but got some errors about
> {{org.apache.sysml.runtime.DMLRuntimeException: Not supported: Instructions
> of type other than CP instructions}} using the mode {{SPARK}}, and the error
> {{java.lang.NullPointerException}} using the mode {{HYBRID_SPARK}}. More log
> information can be found at the following comments.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
