[Darshan-users] DASK and darshan

Snyder, Shane ssnyder at mcs.anl.gov
Wed Oct 14 08:22:13 CDT 2020


You will also have to set the DARSHAN_ENABLE_NONMPI environment variable (i.e., export DARSHAN_ENABLE_NONMPI=1) to ensure logs are generated in the case of non-MPI applications. As long as that variable is set, you'll get Darshan logs for basically any process running from your shell, so make sure you only set it when you really want instrumentation.

Please let us know how it goes. I've had success getting Darshan working with Spark, and I know other users have had success using with TensorFlow, so hopefully Dask isn't much of a problem.

--Shane
________________________________
From: Darshan-users <darshan-users-bounces at lists.mcs.anl.gov> on behalf of Harms, Kevin <harms at alcf.anl.gov>
Sent: Tuesday, October 13, 2020 6:53 PM
To: Razvan Stefanescu <razvan.stefanescu at spire.com>; darshan-users at lists.mcs.anl.gov <darshan-users at lists.mcs.anl.gov>
Subject: Re: [Darshan-users] DASK and darshan

Razvan,

  It's be a couple of years since i used Dask, but you should be able to profile by setting LD_PRELOAD=/path/to/darshan/install/lib/libdarshan.so.
  Please note that Dask will start a process for each python instance so you will have a darshan log for each process. A few more caveats, if you're using remote processes, I'm not sure if the Dask servers will pass environment variables or not. I think Dask has other non-distributed schedulers that might work differently if running on a single node.

  Also note, by default Darshan assumes MPI and depends on it. If you're applications aren't using MPI, then make sure you configure darshan with the switch --without-mpi.

kevin

________________________________________
From: Darshan-users <darshan-users-bounces at lists.mcs.anl.gov> on behalf of Razvan Stefanescu <razvan.stefanescu at spire.com>
Sent: Tuesday, October 13, 2020 4:41 PM
To: darshan-users at lists.mcs.anl.gov
Subject: [Darshan-users] DASK and darshan

Hello All,

We need to profile some DASK parallel I/O application, and I wonder if you can provide an example on how to use darshan to perform such analysis. Your help is much appreciated.

Thank you,

Razvan

--
RAZVAN STEFANESCU
Head of Statistics and Machine Learning Branch

Senior Data Assimilation and Data Scientist

Spire Global, Inc.

1050 Walnut Street, Suite 402, Boulder, CO 80302 USA

+1-720-<tel:7206432245>643-2231

+1-850-443-1718<tel:4178491948>
_______________________________________________
Darshan-users mailing list
Darshan-users at lists.mcs.anl.gov
https://lists.mcs.anl.gov/mailman/listinfo/darshan-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/darshan-users/attachments/20201014/ad3fb708/attachment-0001.html>


More information about the Darshan-users mailing list