Hum… I used the ipcluster command on the vis node I’m using

And I get through all your init steps:

In [1]:

from IPython.parallel import Client

rc = Client()

In [2]:

rc.ids

Out[2]:

[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]

In [3]:

%autopx

%autopx enabled

In [4]:

from mpi4py import MPI

 

MPI.COMM_WORLD.Get_rank()

 

Out[0:1]: 8

Out[1:1]: 1

Out[2:1]: 14

Out[3:1]: 7

Out[4:1]: 9

Out[5:1]: 15

Out[6:1]: 6

Out[7:1]: 12

Out[8:1]: 4

Out[9:1]: 0

Out[10:1]: 5

Out[11:1]: 10

Out[12:1]: 2

Out[13:1]: 11

Out[14:1]: 3

Out[15:1]: 13

In [5]:

But then it can’t find yt!! 

import yt

yt.enable_parallelism()

import numpy as np

import os

import shutil

import h5py

import gc

import glob

AttributeError                            Traceback (most recent call last)<ipython-input-3-915313e529f2> in <module>()
----> 1 import yt

...

Weird tho … cuz if I kill the cluster and try again, import YT loads… Can you think of anything obvious? It’s like my path to yt goes away once I start the cluster … and I’m getting some default version (old) on stampede.

Thanks Nathan .. I’m watching the tutorial now … 

Cheers!

Rick Sarmento

SESE Astronomy/Astrophysics Grad Student

rsarment@asu.edu

On Jan 17, 2015, at 12:54 PM, Rick Sarmento <rsarment@asu.edu> wrote:

Thanks for the info Nathan… 

I’m trying to use STAMPEDE out at Texas (http://vis.tacc.xsede.org/), and they have a way to kick-off access to their 16 core visualization nodes via the web… But that probably doesn’t help given the info below. I’ll try to work with support to see how I can get it to work… 

If I start the cluster using your instructions (via a ssh session), below, I’d have to login — but then I can’t get remote access to the notebook interface because of the firewalls, etc. Although there’s like a way to setup an ssh tunnel, I haven’t figured it out yet.

Thx!

Cheers!

Rick Sarmento

SESE Astronomy/Astrophysics Grad Student

rsarment@asu.edu

Hi Rick,

You need to launch the notebook in parallel using an MPI parallel IPython cluster.  In addition, you will need to configure the notebook to hook into IPython's built-in parallelism.

I've uploaded a notebook that uses parallel IPython here:

http://nbviewer.ipython.org/gist/ngoldbaum/32ca07cf4f5b3dd06add

Note that all the yt operations aren't really important, but the first few cells where I check to make sure IPython's parallelism works and that MPI parllelism is working are quite important.

Note also that you will need to launch the IPython cluster separately from the notebook server using the "ipcluster" command.

If you want to learn more about parallel IPython, I'd encourage you to take a look at Min Ranger-Kelley's tutorial from scipy 2014: http://pyvideo.org/video/2738/interactive-parallel-computing-with-ipython-part

All that said, it does add a significant amount of semantic overhead to use the IPython notebook in parallel.  It's generally much more straightforward to work with yt in parallel using regular python scripts.

Hope that helps,

Nathan