remote pycallgraph

Author: edbennett

_episodes/05-profiling.md

@@ -285,80 +285,6 @@ speed by using Pathos rather than GNU Parallel—see this
 aftenroon's session.
 
 
-## Using `pycallgraph` to see a flowchart of function calls
-
-SnakeViz gives a very useful graphical representation of how different
-functions' execution times relate to each other. In many cases it can
-also show which functions call which, but in more complex cases this
-becomes unclear. `pycallgraph` gives an alternative view of this; it
-doesn't use `cProfile`, but instead runs the program itself and
-records which functions call which, and how long they take. A
-disadvantage of this over `cProfile` is that as it is written in pure
-Python, it adds significant overhead to the program's run time.
-
-To get started with `pycallgraph` we will need to install it into our
-conda environment on the cluster. In addition, we also need to install
-Graphviz, a piece of software for rendering graphs which `pycallgraph`
-depends on.
-
-~~~
-$ pip install pycallgraph
-$ conda install graphviz
-~~~
-{: .language-bash}
-
-With that out of the way, we can re-run the Monte Carlo example, now
-using `pycallgraph`. Similarly to `cProfile`, there is an initial part
-telling `pycallgraph` what to do, followed by the Python program to
-run and any arguments.
-
-~~~
-$ pycallgraph graphviz -- ./mc.py 0 1 0.1 1000000 test.dat
-~~~
-{: .language-bash}
-
-Once this completes, it will create a file called `pycallgraph.png` in
-the current working directory. You can copy this to your machine to
-view using FileZilla, or with the command
-
-~~~
-$ # This runs on your computer, not on the supercomputer
-$ scp s.your.username@sunbird.swansea.ac.uk:high-performance-python-examples/pycallgraph.png ~/Desktop/
-~~~
-{: .language-bash}
-
-Open this image in your favourite image viewer, and you can get a look
-at how the function calls in the program relate to each
-other. Functions that are part of Python are in grey, while functions
-that are part of your program are in more interesting colours.
-
-> ## Displaying graphics from the cluster
->
-> If you have a piece of software called an "X Server" installed on
-> your machine, you can display graphics from the cluster without
-> having to copy image files back. GNU/Linux's graphics are based on
-> X, so if you are running GNU/Linux then you won't need to install
-> anything. On Mac, [XQuartz](https://www.xquartz.org) will do this
-> job, while on Windows, 
-> [VcXsrv](https://sourceforge.net/projects/vcxsrv/) is available.
->
-> Once this is installed, you can add the `-X` flag to your `ssh`
-> command when connecting to the cluster, and graphical applications
-> running on the cluster will appear on your machine. For example, to
-> display the output of `pycallgraph`, you can run
->
-> ~~~
-> display pycallgraph.png
-> ~~~
-> {:.language-bash}
-{: .callout}
-
-The main result we can see from this program is that `pycallgraph` has
-increased the run time by an order of magnitude compared to
-`cProfile`; this program isn't complicated enough to benefit from this
-detailed view of the call graph.
-
-
 ## Using `timeit` for quick and dirty timings
 
 Sometimes you don't need a full profile; all you're interested in is