A practical solution to massive parallel computing in R with caret machine learning and foreach examples
Go to file
Andrzej Wójtowicz a57b7f529f changed swapon procedure;
added disabling swap
2016-11-26 15:48:22 +01:00
.gitignore Updated bash script and gitignore 2016-11-08 16:21:28 +01:00
README.md Update README.md 2016-11-17 12:38:06 +01:00
config-parallel.R added renice, changing passwords, setting poweroff; 2016-11-17 12:26:44 +01:00
config.R Added core files 2016-11-08 13:49:28 +01:00
init-parallel.R added renice, changing passwords, setting poweroff; 2016-11-17 12:26:44 +01:00
init.R Added core files 2016-11-08 13:49:28 +01:00
r-distributed-computing.Rproj Added core files 2016-11-08 13:49:28 +01:00
remote-commands.sh changed swapon procedure; 2016-11-26 15:48:22 +01:00
test.R Added core files 2016-11-08 13:49:28 +01:00
utils-parallel.R added renice, changing passwords, setting poweroff; 2016-11-17 12:26:44 +01:00
utils.R added renice, changing passwords, setting poweroff; 2016-11-17 12:26:44 +01:00

README.md

R distributed computing

In this repository I show a practical solution to massive distributed computing in R. I have successfully tested this approach in my research with ~220 remote hosts (and ~800 CPUs) on computers located in Faculty of Mathematics and Computer Science, Adam Mickiewicz University in Poznań. Brief examples concern caret package for creating predictive models and foreach loop for more general parallel computing.

The solution is based on:

Quick example

Suppose you have three Debian-based machines with at least 4 GB RAM:

  1. server (192.168.0.1),
  2. two dual-core hosts (192.168.0.2 and 192.168.0.3; WMI rescue - small Linux image based on Debian).

On the server you install necessary packages and R with project libraries:

[~/r-distributed-computing]$ sudo bash remote-commands.sh install_env install_mro
[~/r-distributed-computing]$ Rscript init.R

Then you prepare remote hosts:

[~/r-distributed-computing]$ echo "192.168.0.2
192.168.0.3" | bash remote-commands.sh configure_hosts

Finally, you run your calculations:

[~/r-distributed-computing]$ Rscript test.R
Scanning for packages used in this project
  |====================================================================| 100%
- Discovered 14 packages
All detected packages already installed
checkpoint process complete
---
INFO [2016-11-08 13:30:46] Creating remote PSOCK cluster
INFO [2016-11-08 13:30:46] [1/4] Connecting to 192.168.0.2 ... 
INFO [2016-11-08 13:30:46] OK
INFO [2016-11-08 13:30:46] [2/4] Connecting to 192.168.0.2 ... 
INFO [2016-11-08 13:30:46] OK
INFO [2016-11-08 13:30:46] [3/4] Connecting to 192.168.0.3 ... 
INFO [2016-11-08 13:30:47] OK
INFO [2016-11-08 13:30:46] [4/4] Connecting to 192.168.0.3 ... 
INFO [2016-11-08 13:30:47] OK
INFO [2016-11-08 13:30:47] Working on 4 nodes
INFO [2016-11-08 13:30:47] Exporting checkpoint constants
INFO [2016-11-08 13:30:54] Setting cluster RNG kind
INFO [2016-11-08 13:30:54] Registering cluster
INFO [2016-11-08 13:30:54] *************************
INFO [2016-11-08 13:30:54] Test foreach
[1] 1 2 3 4
INFO [2016-11-08 13:30:55] Test caret
Support Vector Machines with Linear Kernel 

32 samples
10 predictors

No pre-processing
Resampling: Cross-Validated (10 fold, repeated 10 times) 
Summary of sample sizes: 28, 29, 29, 29, 28, 29, ... 
Resampling results:

  RMSE      Rsquared 
  3.276949  0.8328294

Tuning parameter 'C' was held constant at a value of 1

INFO [2016-11-08 13:31:54] Workers shut down

Customization

Bash control script

The bash script remote-commands.sh is responsible for the server and hosts preparation for calculations. The script reads list of remote hosts from either stdin or file set in CONNECTION_LIST_FILE script variable (default: remote-connection-list.txt).

Then the script executes internal procedures; the names and execution order of such functions are passed as script arguments, e.g. bash remote-commands.sh configure_hosts.

The configure_hosts is a short name for basic execution order:

  1. generate_ssh_keys - generates ssh keys to communicate with hosts (see SSH_* variables),
  2. hosts_push_ssh_key - pushes the keys to hosts and disables password authentication,
  3. hosts_change_password - changes on hosts default password to user-defined in NEW_PASS,
  4. hosts_push_shell_script - pushes the bash script to hosts,
  5. hosts_enable_swap - enables on hosts swap defined in SWAP_PART,
  6. dump_project_r_files - gathers all R project files used in the project,
  7. dump_r_libraries - gathers all R libraries in ~/.checkpoint directory,
  8. hosts_push_project_r_files - pushes gathered R project files to hosts,
  9. hosts_install_env - updates and installs packages defined in DEBIAN_PACKAGES_TO_INSTALL,
  10. hosts_set_power_off - sets power off time as defined in POWEROFF_TIME,
  11. hosts_install_mro - installs R defined in MRO_* variables,
  12. hosts_push_r_libraries_dump - pushes R project files to hosts,
  13. make_remote_connection_list_nproc - creates connection list file for R, defined in HOSTS_FILE (default: remote-hosts.txt).

You can also scan and limit hosts to those currently available through hosts_scan_available procedure and check if there are at least MIN_HOSTS available hosts.

Instead of pushing to hosts precompiled R libraries (hosts_push_r_libraries_dump) you can force compiling them on the hosts (hosts_install_r_libraries).

You may customize MRO_INSTALL_URL to your own mirror - from my experience the MRAN servers limit the download bandwidth in case.

If you want to make only one connection per node (regardless number of cores), you can execute make_remote_connection_list_single instead of make_remote_connection_list_nproc. You may also try to detect logical cores through setting REMOTE_DETECT_LOGICAL_CPUS.

Alternatively, instead of installing soft on WMI rescue, you can create and boot on hosts your own customized distro.

You can investigate hosts install logs through hosts_check_install_log_* functions. You can also check and clean remote worker logs through hosts_check_worker_log and hosts_clean_worker_log procedures, respectively. Moreover, you can investigate dmesg through hosts_check_worker_dmesg. The hosts can be invoked to power off through hosts_power_off.

The functions (excluding hosts_scan_available) stop the script if any part of the procedure fails.

R project files

In the config.R file you may customize checkpoint library configuration and the logging system.

In the config-parallel.R you can:

  • switch between parallel computing methods (PARALLEL.USED.METHOD by default is REMOTE but for testing you can change it to LOCAL, so that all calculations will be done on the server); for LOCAL the method defined in PARALLEL.LOCAL.METHOD might be PSOCK (default) or FORK,
  • set or disable priority for the worker processes through PARALLEL.RENICE variable,
  • customize remote R session invoking parameters through PARALLEL.REMOTE.* variables,
  • worker logs file names are defined in PARALLEL.*.SLAVE.OUT.FILE and are unique among whole computing cluster.

Working with more than 125 connections

Currently R has a hard-coded limit for number of connections, which is effectively 125. If you want to work with more connections (nodes), then on the server you have to recompile and install customized R. You may see how to set custom limit in my repository Microsoft R Open compilation for customized cluster nodes connection limit.

Reproducibility

In caret you can control reproducibility by pre-setting seeds; see section 5.4 Notes on Reproducibility.

Intel MKL

By default the Intel MKL support is installed with Microsoft R Open. In config-parallel.R you can choose in PARALLEL.DISABLE.MKL.THREADS variable either this library should be either disabled (default) or enabled. This can be useful in particular if you make cluster in the one-host-one-connection manner and calculations on remote nodes can benefit from BLAS routines.