José Abell's research blog

Useful Commands

Useful Commands for everyday tasks

Table of Contents


Linux Management

To use find to remove files

find . -name "*.o" -exec rm {} \;

To find and delete files matching a criteria but not another

find . -depth -name "*.pdf" -prune -not -name "*graded.pdf" -delete


find . -name "*.cpp" -print | xargs sed -i 's/jose/jaabell/g'

Find recently changed files

find -mtime 0 -type f -print

Find where apt-get installs things

sudo dpkg -L packagename

Killing a screen session

screen -X -S [session # you want to kill] quit

Deleting lines from file matching a criteria

sed '/pattern to match/d' ./infile

Control volume from command line

amixer -D pulse sset Master 100%

Find and delete all broken symlinks in a folder

find . -xtype l -delete

List folders by space utilized

du -hs * | sort -h


Setting IP address from Command Line:

sudo ifconfig eth0 netmask

Linux system setup


Don’t use gnome app, simply use


View printing queue from command line


Cancel a job


Encrypt and decrypt files

  openssl enc -aes-256-cbc -e > out.tar.gz.enc
  openssl enc aes-256-cbc -d -in out.tar.gz.enc | tar xz

Git Management

Create and checkout a local branch and set it up to track a remote branch

git checkout -b master -t origin/master

Where origin/master is a remote branch.

Git history cleanup:

git gc --aggressive

Adding a branch

git push <remote-name> <branch-name>

Formally git push : But when you omit one, it assumes both names are the same. Do:

git push -u <remote-name> <branch-name>

So that a subsequent git pull will know what to do (set upstream tracking)

Setting up a non-empty folder to be a git repo

cd <localdir>
git init
git add .
git commit -m 'message'
git remote add origin <url>
git push -u origin master

Setting up upstream in GitHub

git remote add upstream


git remote -v

SSH Tricks

Shutdown a frozen SSH session. Press :


Proceed to resume happyness.

Tunnelling all traffic through SSH tunnel

Use sshuttle.

sshuttle -r essi 0/0 -vv

Managing gitolite

Checkout the gitolite admin repo:

git clone git@powa:gitolite-admin gadmin

To add a user SSH key do the following:

  • Add it to the keydir
  • Multiple keys for same user either in subfolders or using @ ie. equivalently

    • home/
  • Stage it and commit it

    git add . 
    git commit -m "message"
  • Push to the gitolite repo

    git push

Changing admin settings

  • Open the file gitolite-admin/cong/gitolite.conf

  • Make changes, commit and push

    git commit -am "message"
    git push

If directly changing stuff in the gitolite server (ie. logging in through ssh into the git user)

  • Remember to use:

    ssh (myusername)
    su git  [input the typical CompGeoMech password]
  • Make some changes and then


To add a new repo:

  • Add a new line into the conf/gitolite.conf file
  • Push changes into gitolite
  • Pull the new (empty) repo
  • Add stuff to it and push
  • Good to go!

Python cool stuff

Iteration awesomeness.


a = ['a1', 'a2', 'a3']
b = ['b1', 'b2']

will iterate 3 times, the last iteration, b will be None.

print "Map:"
for x, y in map(None, a, b):
  print x, y

will iterate 2 times, the third value of a will not be used

print "Zip:"
for x, y in zip(a, b):
  print x, y

will iterate 6 times, it will iterate over each b, for each a producing a slightly different output

print "List:"
for x, y in [(x,y) for x in a for y in b]:
    print x, y

IPython Notebook

Install pip

sudo apt-get install python-pip
sudo pip install ipython --upgrade
sudo pip install ipython[notebook]

IPython Notebook server


sudo apt-get install ipython-notebook


from IPython.lib import passwd

#Enter passwd

Create an SSL certificate to encrypt session openssl req -x509 -nodes -days 365 -newkey rsa:1024 -keyout mycert.pem -out mycert.pem

Edit to look something like this

c = get_config()
c.NotebookApp.open_browser = False
c.NotebookApp.port = 16562
c.NotebookApp.certfile = u'/home/jaabell/ipython_notebook_ssl_certificate.pem'
c.NotebookApp.password = u'sha1:e6356b6e0ee8:1906d7f56995bc18925c4af976ac3a588bc10913'
c.IPKernelApp.ip = u'*'
c.IPKernelApp.pylab = 'inline'
c.FileNotebookManager.notebook_dir = u'/home/jaabell/www/blog/content/ipython_notebooks'

Create a profile for the server

ipython profile create nbserver

Edit ~/.config/ipython/profile_nbserver/

Run the server

 ipython notebook --profile=nbserver

Keep the server alive with supervisor

sudo apt-get install supervisor


To view contents of a static library (*.a)

nm -C <PATH_TO_LIB>.a > contents.txt

The -C flag is for C++.

To search a bunch of libraries for a specific function:

for lib in $(find $PATH -name \*.a) ; do echo $lib ; nm $lib | grep "specific_function" | grep -v " U "   ; done


After running a program compile with -g -pg the call graph can be visualized with and xdot (or dot).

gprof ~/bin/essi_semidebug | | dot -Tdot -o

or just create a png

gprof ~/bin/essi_semidebug | | dot -Tpng -o output.png


Compile MVAPICH2 with message queue debugging.

./configure --prefix=".." --enable-debuginfo --disable-mcast

Get libnuma to do optimization for NUMA architechtures.

sudo apt-get install libnuma-dev


To use a different compiler with mpicc and mpic++ export OMPI_CC=clang export OMPI_CPP=clang++

MPI Performance


Need to try this out. Its supposed to be good.


Get from

Requires libunwind, binutils and libiberty

sudo apt-get install libunwind8 libunwind8-dev binutils-dev libiberty-dev


./configure --enable-collective-report-default --enable-demangling=GNU --with-cc=mpicc --with-cxx=mpic++ --with-f77=mpif77 --prefix=/usr/local 
make -j 24
sudo make install

If all good ll /usr/local/lib/libmp* should return /usr/local/lib/libmpiP.a

Append to code at linktime the following flags

-lmpiP -lbfd -liberty -lunwind

Possibly also the place where libmpiP.a is available throught the -L option. Ie. for above -L/usr/local… unnecesarry.


Monitor and profile performance counters through PAPI interaface. Has nice viewer, apparently.

Building and installation

From Checkout both hpctoolkit and hpctoolkit-externalsL:

svn co hpctoolkit
svn co hpctoolkit-externals

Build externals (prefix set to a local dir):

cd hpctoolkit-externals
mkdir BUILD && cd BUILD
../configure CC=gcc CXX=g++ --prefix=/home/jaabell/Programs/local/
make -j 8
make install
make clean

Build the toolkit. On ubuntu 14.04 libpapi is located on /usr/lib/x86_64-linux-gnu, make links into /usr/lib for this to work.

cd hpctoolkit
mkdir BUILD && cd BUILD
../configure CC=gcc CXX=g++ MPICXX=mpic++ --prefix=/usr/local --with-externals=/home/jaabell/Programs/local --with-papi=/usr/
make -j 8
sudo make install -j 8


Figure out list of PAPI events:


Some interesting ones for me, that were available on my laptop

PAPI_L1_DCM  : Level 1 data cache misses
PAPI_L2_DCM  : Level 2 data cache misses
PAPI_L3_DCM  : Level 3 data cache misses
PAPI_TLB_DM  : Data translation lookaside buffer misses
PAPI_FP_INS  : Floating point instructions
PAPI_TOT_CYC : Total cycles
PAPI_FP_OPS  : Floating point operations
PAPI_VEC_DP  : Double precision vector/SIMD instructions

Can also get a list of events (PAPI and native) using -L option

hpcrun -L essi > HPCeventlist.txt

Run application through hpcrun (this worked on nagoy)

hpcrun \
-e PAPI_L1_DCM@5300013 \
-e PAPI_L2_DCM@5300013 \
-e PAPI_TOT_CYC@5300013 \
-e PAPI_FP_OPS@5300013 \
essi -nf <file>


Provides a simple command-line interface into performance counters.

Get from


Requires PAPI library libpapi.a to be in a reasonable locations. See HPCToolkit.

./configure --prefix=/usr/local/ --with-papi=/usr
make -j 16
sudo make install


Tuning BLAS and LAPACK with ATLAS



Install cpufrequtils:

sudo apt-get install cpufrequtils

Then edit the following file (if it doesn’t exist, create it):

sudo nano /etc/default/cpufrequtils

And add the following line to it:


Save and exit.

If you need performance, do:

sudo /etc/init.d/cpufrequtils restart

Finding out “real” CPUS from hyperthreading ones

On my laptop Intel(R) Core(TM) i7-2630QM CPU @ 2.00GHz

Look at cat /proc/cpuinfo and look at the core ids. Pick processors which are on different cores.

--force-tids="4 0 1 2 3"


Now to tune BLAS and LAPACK with ATLAS….

mv ATLAS ATLAS3.10.x
cd ATLAS3.10.x
mkdir linux_x86_64_intel_i7_laptop
cd linux_x86_64_intel_i7_laptop
../configure -b 64 -D c -DPentiumCPS=2000 --prefix=/home/jaabell/Programs/lib  --with-netlib-lapack-tarfile=/home/jaabell/Programs/lapack-3.5.0.tgz --force-tids="4 0 1 2 3"

Explanation -b is the pointer bitwidth, -D c -DPentiumCPS=2000 sets the CPU clock rate so that ATLAS can use CPU cycles for timing, --prefix where to install, -with-netlib-lapack-tarfile= where is the lapack tarball, and --force-tids="4 0 1 2 3" tells ATLAS to only use those core ids given (first number is the number of cores to use).

make build   # tune & build lib
make check   # sanity check correct answer
make ptcheck # sanity check parallel
make time    # check if lib is fast
make install # copy libs to install dir



Edit I added:

Compilers… CC = gcc CXX = g++ TARGET_ARCH = -march=native AR = gcc-ar

Installation… INSTALL_LIB = /home/jaabell/Programs/lib INSTALL_INCLUDE = /home/jaabell/Programs/include

BLAS and LAPACK come from ATLAS… ESSI_DEPEND_DIR = /home/jaabell/Repositories/essi_dependencies BLAS = $(ESSI_DEPEND_DIR)/lib/libcblas.a $(ESSI_DEPEND_DIR)/lib/libf77blas.a $(ESSI_DEPEND_DIR)/lib/libatlas.a -lgfortran LAPACK = $(ESSI_DEPEND_DIR)/lib/liblapack.a

Other cool stuff

Making videos of stuff

Screen capture / streaming / making desktop videos with audio

  • on bin/recordscreen (python script)

    see recordscreen —help

  • kazam (sudo apt-get install kazam)

Making a movie from separate png images.


avconv -r 60 -i movie%04d.png  movie.mp4

The options are

  • qscale 5seems to be deprecated define fixed video quantizer scale (VBR) where 1 is the best and 31 the worst. Since mpeg/jpeg has problems to compress line graphics it’s a good idea to set this variable close to 1. You get a big movie file, but otherwise the movie doesn’t look, well, that good.
  • r: framerate
  • b: video bitrate
  • i: input files, %04d says that we have four numbers in the filename where the number is filled with zeros left of it. movie.mp4 is the filename, the extension says that it is a quicktime movie. You can also create a Macromedia Flash movie by using the .flv extension.

to get higher quality (play with crf parameter = 1 (best) = 31 (worst))

avconv -r 60 -i movie%04d.png -c:v libx264 -crf 2 ../../movie_2_npps.mp4

for non-even sized movies a little trick should be played:

avconv -r 60 -i movie%04d.png  -vf "scale=trunc(iw/2)*2:trunc(ih/2)*2" movie.mp4

Explanation for -vf parameter, from StackOverflow

Basically the issue stems from a bug(?) in libx264 where it complains if the width or height is not an even number. This is weird in the case where I don’t want to perform any scaling at all. So the command above will:

  • Divide the original height and width by 2
  • Round it down to the nearest pixel
  • Multiply it by 2 again, thus making it an even number

To get it to work on adobe reader use h264 encoder!

avconv -i source -c:v h264 -c:a copy out.mkv
avconv -i source -c:v libx264 -c:a copy out.mkv

To stick together two videos side by side

avconv -i movie_left.mp4 -vf "[in] scale=trunc(iw/2)*2:trunc(ih/2)*2, pad=2*iw:ih [left];     movie=movie_right.mp4, scale=trunc(iw/2)*2:trunc(ih/2)*2 [right];     [left][right] overlay=main_w/2:0 [out]"  Output.mp4


On Ubuntu gdb attaching to processes is forbidden… because hackers. To allow:

echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope

To permanently allow it edit /etc/sysctl.d/10-ptrace.conf and change the line:

kernel.yama.ptrace_scope = 1

To read

kernel.yama.ptrace_scope = 0

Website Management (with pelican)

Install pelican using tutorials in website.

Setup: sudo apt-get install python-pip sudo pip install pelican fabric sudo pip install pelican markdown typogrify pelican_youtube ipython

Use fabric or make to build

Using vitualenv

virtualenv ~/virtualenvs/pelican
cd ~/virtualenvs/pelican
. bin/activate

Download the [ipythonnb plugin][]

git clone

Will need pandoc

sudo apt-get install pandoc pandoc-citeproc

Apache management

Once apache is running

To create a new website

  • Create a new virtual host inside sites-available. Here is the code for my website:

    ServerAdmin ServerName DocumentRoot /var/www/ Options FollowSymLinks AllowOverride None Options Indexes FollowSymLinks MultiViews AllowOverride None Order allow,deny allow from all

  • Link from sites-available to sites-enabled

  • Restart Apache

    /etc/init.d/apache2 restart

Tweaks to .bashrc

Add screen name to prompt

#For screen
if [ "$TERM" = screen ]; then

Make terminals solarized

wget --no-check-certificate dark
mv dircolors.ansi-dark .dircolors
eval `dircolors ~/.dircolors`

wget --no-check-certificate light
mv dircolors.ansi-light .dircolors
eval `dircolors ~/.dircolors`

sudo apt-get install git-core
git clone
cd gnome-terminal-colors-solarized


VisIt tricks

Make window a given size

Save a session file, and edit these lines to achieve desired res:

<Field name="windowSize" type="intArray" length="2">1024 828 </Field>
<Field name="windowImageSize" type="intArray" length="2">1024 768 </Field>
<Field name="windowLocation" type="intArray" length="2">719 133 </Field>

HDF5 Tricks

Dump part of an array

h5dump can be used to output portions of an HDF5 file into text, binary or other HDF5 files.

Example 1 Dump 1000 timesteps of the first row of generalized displacements in FILENAME.h5.feioutput.

h5dump -d "/Model/Nodes/Generalized_Displacements" --start="0,0" --count="1,1000" FILENAME.h5.feioutput

Repacking an HDF5 file

Repacking is useful to change chunking layout or add filters (such as compression) after the file has been created.

Example 1 Change the chunking for the generalized displacements dataset to be 10x20. Do:

h5repack -v -l "/Model/Nodes/Generalized_Displacements":CHUNK=10x20 FILENAME_IN.h5.feioutput FILENAME_OUT.h5.feioutput

Example 2 Apply the best compression (GZIP) to all arrays in a file:

h5repack -v -f GZIP=9 FILENAME_IN.h5.feioutput  FILENAME_IN.h5.gzip.feioutput