Many users tell me that R is slow. With old R releases that is 100% true provided old R versions used its own numerical libraries instead of optimized numerical libraries.
But, numerical libraries do not explain the complete story. In many cases slow code execution can be attributed to inefficient code and in precise terms because of not doing one or more of these good practises:
- Using byte-code compiler
- Vectorizing operations
- Using simple data structures (i.e using data frames instead of matrices in large computing instances)
- Re-using results
I would add another good practise: “Use the tidyverse”. Provided tidyverse packages such as
dplyr benefit from
Rcpp, having a C++ backend can be faster than using dplyr’s equivalents in base (i.e plain vanilla) R.
The idea of this post is to clarify some ideas. R does not compete with C or C++ provided they are different languages. R and
data.table package may compete with Python and
numpy library. This does not mean that I’m defending R over Python or backwards. The reason behind this is that both R and Python implementations consists in an interpreter while in C and C++ it consists in a compiler, and this means that C and C++ will always be faster because in really over-simplifying terms compiler implementations are closer to the machine.
Basic setup for general usage
As an Ubuntu user I can say the basic R installation from Canonical or CRAN repositories work for most of the things I do on my laptop.
When I use RStudio Server Pro© that’s a different story because I really want to optimize things because when I work with large data (i.e. 100GB in RAM) a 3% more of resources efficiency or reduced execution time is valuable.
Installing R with OpenBLAS will give you a tremedous performance boost, and that will work for most of laptop situations. I explain how to do that in detail for Ubuntu 17.10 and Ubuntu 16.04 but a general setup would be as simple as one of this two options:
# 1: Install from Canonical (default Ubuntu repository) sudo apt-get update && sudo apt-get upgrade sudo apt-get install libopenblas-dev r-base # 2: Install from CRAN mirror sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 51716619E084DAB9 printf '#CRAN mirror\ndeb https://cran.rstudio.com/bin/linux/ubuntu artful/\ndeb-src https://cran.rstudio.com/bin/linux/ubuntu artful/\n' | sudo tee -a /etc/apt/sources.list.d/cran-mirror.list sudo apt-get update && sudo apt-get upgrade sudo apt-get install libopenblas-dev r-base # 3: Install RStudio (bonus) cd ~/Downloads wget https://download1.rstudio.org/rstudio-xenial-1.1.383-amd64.deb sudo apt-get install gdebi sudo gdebi rstudio-xenial-1.1.383-amd64.deb printf '\nexport QT_STYLE_OVERRIDE=gtk\n' | sudo tee -a ~/.profile
Being (1) a substitute of (2). It’s totally up to you which one to use and both will give you a really fast R compared to installing it without OpenBLAS.
Benchmarking different R setups
I already use R with OpenBLAS just like the setup above. I will compile parallel R instances to do the benchmarking.
Installing Intel© MKL numerical libraries
My benchmarks do indicate that in my case it’s convenient to take the time it takes to install Intel© MKL. The execution time is strongly reduces for some operations when compared to R with OpenBLAS performance.
Run this to install MKL:
# keys taken from https://software.intel.com/en-us/articles/installing-intel-free-libs-and-python-apt-repo wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB sudo sh -c 'echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list' sudo apt-get update && sudo apt-get install intel-mkl-64bit
Installing CRAN R with MKL
To compile it from source (in this case it’s the only option) run these lines:
# key added after sudo apt-get update returned a warning following this guide: https://support.rstudio.com/hc/en-us/articles/218004217-Building-R-from-source sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 51716619E084DAB9 printf '#CRAN mirror\ndeb https://cran.rstudio.com/bin/linux/ubuntu artful/\ndeb-src https://cran.rstudio.com/bin/linux/ubuntu artful/\n' | sudo tee -a /etc/apt/sources.list.d/cran-mirror.list # you need to enable multiverse repo or packages as xvfb won't be found sudo rm -rf /etc/apt/sources.list printf 'deb http://us.archive.ubuntu.com/ubuntu artful main restricted universe multiverse deb-src http://us.archive.ubuntu.com/ubuntu artful main restricted universe multiverse\n deb http://security.ubuntu.com/ubuntu artful-security main restricted universe multiverse deb-src http://security.ubuntu.com/ubuntu artful-security main restricted universe multiverse\n deb http://us.archive.ubuntu.com/ubuntu artful-updates main restricted universe multiverse deb-src http://us.archive.ubuntu.com/ubuntu artful-updates main restricted universe multiverse\n' | sudo tee -a /etc/apt/sources.list sudo apt-get update sudo apt-get clean sudo apt-get autoclean sudo apt-get autoremove sudo apt-get upgrade --with-new-pkgs sudo apt-get build-dep r-base cd ~/GitHub/r-with-intel-mkl wget https://cran.r-project.org/src/base/R-3/R-3.4.2.tar.gz tar xzvf R-3.4.2.tar.gz cd R-3.4.2 source /opt/intel/mkl/bin/mklvars.sh intel64 MKL="-Wl,--no-as-needed -lmkl_gf_lp64 -Wl,--start-group -lmkl_gnu_thread -lmkl_core -Wl,--end-group -fopenmp -ldl -lpthread -lm" ./configure --prefix=/opt/R/R-3.4.2-intel-mkl --enable-R-shlib --with-blas="$MKL" --with-lapack make && sudo make install printf '\nexport RSTUDIO_WHICH_R=/usr/local/bin/R\nexport RSTUDIO_WHICH_R=/opt/R/R-3.4.2-intel-mkl\n' | tee -a ~/.profile
Installing CRAN R with OpenBLAS
Just not to interfere with working installation I decided to compile a parallel instance from source:
cd ~/GitHub/r-with-intel-mkl/ rm -rf R-3.4.2 tar xzvf R-3.4.2.tar.gz cd R-3.4.2 ./configure --prefix=/opt/R/R-3.4.2-openblas --enable-R-shlib --with-blas --with-lapack make && sudo make install printf 'export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-openblas/bin/R\n' | tee -a ~/.profile
Installing CRAN R with no optimized numerical libraries
There is a lot of discussion and strong evidence from different stakeholders in the R community that do indicate that this is by far the most inefficient option. I compiled this just to make a complete benchmark:
cd ~/GitHub/r-with-intel-mkl/ rm -rf R-3.4.2 tar xzvf R-3.4.2.tar.gz cd R-3.4.2 ./configure --prefix=/opt/R/R-3.4.2-defaults --enable-R-shlib make && sudo make install printf 'export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-defaults/bin/R\n' | tee -a ~/.profile
Installing Microsoft© R Open with MKL
This R version includes MKL by default and it’s supposed to be easy to install. I could not make it run and that’s bad because different articles (like this post by Brett Klamer) state that this R version is really efficient but no different to standard CRAN R with MKL numerical libraries.
In any case here’s the code to install this version:
cd ~/GitHub/r-with-intel-mkl wget https://mran.blob.core.windows.net/install/mro/3.4.2/microsoft-r-open-3.4.2.tar.gz tar xzvf microsoft-r-open-3.4.2.tar.gz cd microsoft-r-open sudo ./install.sh printf 'export RSTUDIO_WHICH_R=/opt/microsoft/ropen/3.4.2/lib64/R/bin/R\n' | tee -a ~/.profile # it was not possible to start /opt/microsoft/ropen/3.4.2/lib64/R/bin/R # the error is: # *** caught segfault *** # address 0x50, cause 'memory not mapped' # removing Microsoft R # https://mran.microsoft.com/documents/rro/installation#revorinst-uninstall steps did not work sudo apt-get remove 'microsoft-r-.*' sudo apt-get autoclean && sudo apt-get autoremove
My scripts above do edit
~/.profile. This is to open RStudio and work with differently configured R instances on my computer.
I released the benchmark results and scripts on GitHub. The idea is to run the same scripts from ATT© and Microsoft© to see how different setups perform.
To work with CRAN R with MKL I had to edit
~/.profile because of how I configurated the instances. So I had to run
nano ~/.profile and comment the last part of the file to obtain this result:
#export RSTUDIO_WHICH_R=/usr/bin/R export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-intel-mkl/bin/R #export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-openblas/bin/R #export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-defaults/bin/R #export RSTUDIO_WHICH_R=/opt/microsoft/ropen/3.4.2/lib64/R/bin/R
After that I log out and then log in to open RStudio to run the benchmark.
The other two cases are similar and the benchmark results were obtained editing
~/.profile, logging out and in and opening RStudio with the corresponding instance.
As an example, this result starts with the R version and the corresponding numerical libraries used in that sessions. Any other result are reported in the same way.
And here are the results from ATT© benchmarking script:
|Task||CRAN R with MKL (seconds)||CRAN R with OpenBLAS (seconds)||CRAN R with no optimized libraries (seconds)|
|Creation, transp., deformation of a 2500×2500 matrix (sec)||0.68||0.68||0.67|
|2400×2400 normal distributed random matrix ^1000||0.56||0.56||0.56|
|Sorting of 7,000,000 random values||0.79||0.79||0.79|
|2800×2800 cross-product matrix (b = a’ * a)||0.3||0.36||14.55|
|Linear regr. over a 3000×3000 matrix (c = a \ b’)||0.17||0.22||6.98|
|FFT over 2,400,000 random values||0.33||0.33||0.33|
|Eigenvalues of a 640×640 random matrix||0.22||0.49||0.74|
|Determinant of a 2500×2500 random matrix||0.2||0.22||2.99|
|Cholesky decomposition of a 3000×3000 matrix||0.31||0.21||5.76|
|Inverse of a 1600×1600 random matrix||0.2||0.21||2.79|
|3,500,000 Fibonacci numbers calculation (vector calc)||0.54||0.54||0.54|
|Creation of a 3000×3000 Hilbert matrix (matrix calc)||0.23||0.24||0.23|
|Grand common divisors of 400,000 pairs (recursion)||0.27||0.29||0.3|
|Creation of a 500×500 Toeplitz matrix (loops)||0.28||0.28||0.28|
|Escoufier’s method on a 45×45 matrix (mixed)||0.22||0.23||0.28|
|Total time for all 15 tests||5.3||5.62||37.78|
|Overall mean (weighted mean)||0.31||0.32||0.93|
And here are the results from Microsoft© benchmarking script:
|Task||CRAN R with MKL (seconds)||CRAN R with OpenBLAS (seconds)||CRAN R with no optimized libraries (seconds)|
|Singular Value Decomposition||7.268||18.325||47.076|
|Principal Components Analysis||14.932||40.612||162.338|
|Linear Discriminant Analysis||26.195||43.75||117.537|
Actions after benchmarking results
I decided to try Intel MKL and I’ll write another post benchmarking things I do everyday beyond what is considered in the scripts.
To clean my system I deleted all R instances but MKL:
sudo apt-get remove r-base r-base-dev sudo apt-get remove 'r-cran.*' sudo apt-get autoclean && sudo apt-get autoremove sudo apt-get build-dep r-base sudo rm -rf /opt/R/R-3.4.2-openblas sudo rm -rf /opt/R/R-3.4.2-defaults sudo ln -s /opt/R/R-3.4.2-intel-mkl/bin/R /usr/bin/R
~/.profile so the final lines are:
export RSTUDIO_WHICH_R=/usr/bin/R #export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-intel-mkl/bin/R #export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-openblas/bin/R #export RSTUDIO_WHICH_R=/opt/R/R-3.4.2-defaults/bin/R #export RSTUDIO_WHICH_R=/opt/microsoft/ropen/3.4.2/lib64/R/bin/R
And I also decided to configure my user packages directory from zero:
# remove installed user packages rm -rf ~/R # create new user packages directory mkdir ~/R/ mkdir ~/R/x86_64-pc-linux-gnu-library/ mkdir ~/R/x86_64-pc-linux-gnu-library/3.4 # install common packages R --vanilla << EOF install.packages(c("tidyverse","data.table","dtplyr","devtools","roxygen2","bit64","pacman"), repos = "https://cran.rstudio.com/") q() EOF # Export to HTML/Excel R --vanilla << EOF install.packages(c("htmlTable","openxlsx"), repos = "https://cran.rstudio.com/") q() EOF # Blog tools R --vanilla << EOF install.packages(c("knitr","rmarkdown"), repos='http://cran.us.r-project.org') q() EOF sudo apt-get install python-pip sudo pip install --upgrade --force-reinstall markdown rpy2==2.7.8 pelican==3.6.3 # PDF extraction tools sudo apt-get install libpoppler-cpp-dev default-jre default-jdk sudo R CMD javareconf R --vanilla << EOF library(devtools) install.packages(c("rjava","pdftools"), repos = "https://cran.rstudio.com/") install_github("ropensci/tabulizer") q() EOF # TTF/OTF fonts usage R --vanilla << EOF install.packages("showtext", repos = "https://cran.rstudio.com/") q() EOF # Cairo for graphic devices sudo apt-get install libgtk2.0-dev libxt-dev libcairo2-dev R --vanilla << EOF install.packages("Cairo", repos = "https://cran.rstudio.com/") q() EOF
The benchmark exposed here are in no way a definitive end to the long discussion on numerical libraries. My results show some evidence that indicates, that because of more speed for some operations, I should use MKL.
One of the advantages of the setup I explained is that you can use MKL with Python. In that case
numpy calculations will be boosted.
Using MKL with AMD© processors might not provide an important improvement when compared to use OpenBLAS. This is because MKL uses specific processor instructions that work well with i3 or i5 processors but not neccesarily with non-Intel models.