ݺߣ

ݺߣShare a Scribd company logo

Parallel Computing in R

?
?
mickey24
mickey24

This document discusses using parallel computing in R with the snow package. It provides an overview of using snow to distribute computations across multiple CPUs. Examples are given showing how snow can be used with functions like parApply to speed up matrix multiplication by performing the operation in parallel on a cluster. The document also discusses using snow together with Rmpi and a job scheduler like Sun Grid Engine to enable parallel computing on a computing cluster.

1 of 42
Parallel Computing in R 2010/05/09 Tsukuba.R#7 id:mickey24
id: mickey24 (@mickey24)
Tsukuba.R
Tsukuba.R#4 ?R Brainf*ck ? Brainf*ck > hello <- "+++++++++[>++++++++>++++++++++ +>+++++<<<-]>.>++.+++++++..++ +.>-.------------.<++++++++.--------.++ +.------.--------.>+." > brainfxxk(hello) [1] "Hello, world!" http://www.slideshare.net/mickey24/rbrainfck-1085191
Tsukuba.R#5 ? Animation with R ? library(animation) http://d.hatena.ne.jp/mickey24/20090614
Tsukuba.R#6 ? Extend R with C!!! ? C R (C OpenCV ) http://d.hatena.ne.jp/mickey24/20091123/r_de_extension
Parallel Computing in R
Tsukuba.R#7
? ?n snow ?
n
? ? CPU !"#! !"#! !"#! !!!! !!!! !!!! !"#!
1CPU sapply(1:8, function(x){ x^2 }) (function(x){x^2})(1) (function(x){x^2})(2) (function(x){x^2})(3) (function(x){x^2})(4) !"#! (function(x){x^2})(5) (function(x){x^2})(6) (function(x){x^2})(7) (function(x){x^2})(8) [1] 1 4 9 16 25 36 49 64
4CPU sapply(1:8, function(x){ x^2 }) (function(x){x^2})(1) (function(x){x^2})(3) (function(x){x^2})(5) (function(x){x^2})(7) (function(x){x^2})(2) (function(x){x^2})(4) (function(x){x^2})(6) (function(x){x^2})(8) !"#! !"#! !"#! !"#! [1] 1 4 9 16 25 36 49 64
? ? ? ? 1CPU 1 100CPU 10
R ? snow ? R ? http://cran.r-project.org/web/packages/snow/index.html ? R apply ? ? Socket PVM MPI
n snow
snow ? 2CPU ? ? CPU
? snow > install.packages("snow") ? ?
matprod.R ? 1000 n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) C <- A %*% B ?
clmatprod.R library(snow) n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) cpu <- 2 hosts <- rep("localhost", cpu) cl <- makeCluster(hosts, type="SOCK") C <- parMM(cl, A, B) # C <- A %*% B stopCluster(cl)
? > source("clmatprod.R") ? > head(C)
? ? parMM(cl, A, B) ? apply ? parApply(cl, X, MARGIN, fun, ...) ? parLapply(cl, X, fun, ...) ? parSapply(cl, X, fun, ..., simplify=TRUE, USE.NAMES=TRUE) ? etc. ? ? clusterMap(cl, fun, ..., MoreArgs = NULL, RECYCLE = TRUE) ? clusterCall(cl, fun, ...) ? etc. ? help (?parApply, ?clusterMap )
snow
? ? snow ? mCPU n ? O(n^3) #! $! !"#! !"#! !"#! !!!! !"#! #! !! "!
? 1CPU( ) system.time(A %*% B) ? mCPU ( ) system.time(parMM(cl, A, B))
? @DBCLS ? ? Sun Grid Engine OpenMPI ?
@DBCLS 1 8 CPU 8CPU 64CPU 16GB 128GB !"#$! !"#$! !"#$! !"#$! !"#$! !"#$! !"#$! !"#$!
Parallel Computing in R
1000 ( ) 3.00 2.50 2.00 1.50 1.00 Faster! 0.50 0 1 4 8 16 CPU 4CPU 1.14
3000 ( ) 40.00 30.00 20.00 10.00 Faster! 0 1 4 8 16 CPU 8CPU 2.70
Parallel Computing in R
? snow ? C ? C R (40 50 ) www http://d.hatena.ne.jp/ syou6162/20090117/1232120983
? snow ? snow Simpli?ed http://www.sfu.ca/~sblay/R/snow.html ? RjpWiki - R http://www.okada.jp.org/RWiki/?R%A4%C7%CA%C2%CE%F3%B7%D7%BB%BB ? RjpWiki - L.Tierney snow http://www.okada.jp.org/RWiki/?L.%20Tierney%BB%E1%A4%CEsnow %A5%D1%A5%C3%A5%B1%A1%BC%A5%B8%A4%C7%A5%AF %A5%E9%A5%B9%A5%BF%B7%D7%BB%BB%A4%F2%B9%D4%A4%A6#scce80a1 ? Rmpi + snow + Sun Grid Engine ? Scheduled Parallel Computing with R: R + Rmpi + OpenMPI + Sun Grid Engine (SGE) http://blog.nguyenvq.com/2010/01/20/scheduled-parallel-computing-with-r-r-rmpi- openmpi-sun-grid-engine-sge/
Parallel Computing in R
Parallel Computing in R
R+ Sun Grid Engine + Open MPI + snow
? @DBCLS ? Sun Grid Engine ? OpenMPI ? mpirun R ? Rmpi snow OpenMPI ? ?
? ? Rmpi OpenMPI ? snow Rmpi > install.packages("Rmpi", configure.args= "/path/to/mpidir") > install.packages("snow")
gmatprod.R library(Rmpi) library(snow) n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) cl <- makeCluster() # C <- parMM(cl, A, B) # C <- A %*% B stopCluster(cl)
gmatprod.sh #!/bin/bash #$ -S /bin/bash # /bin/bash #$ -j y # stdout stderr -o #$ -o gmatprod.log # stdout stderr #$ -pe openmpi 8 # export PATH=[path to R & snow_install_dir & mpirun]:$PATH export LD_LIBRARY_PATH=/path/to/mpidir/lib:$LD_LIBRARY_PATH MPIRUN_PATH=/path/to/mpirun MPIRUN_OPTS="-np ${NSLOTS} -machinefile ${TMPDIR}/machines" RSCRIPT=gmatprod.R RPATH=/path/to/snow_install_dir/RMPISNOW CMD="${RPATH} CMD BATCH --no-save ${RSCRIPT}" ${MPIRUN_PATH} ${MPIRUN_OPTS} ${CMD}
? qsub Grid Engine $ qsub gmatprod.sh
? ? Scheduled Parallel Computing with R: R + Rmpi + OpenMPI + Sun Grid Engine (SGE) http://blog.nguyenvq.com/2010/01/20/scheduled-parallel- computing-with-r-r-rmpi-openmpi-sun-grid-engine-sge/ ? Sun Grid Engine MPI R

More Related Content

Parallel Computing in R

  • 1. Parallel Computing in R 2010/05/09 Tsukuba.R#7 id:mickey24
  • 4. Tsukuba.R#4 ?R Brainf*ck ? Brainf*ck > hello <- "+++++++++[>++++++++>++++++++++ +>+++++<<<-]>.>++.+++++++..++ +.>-.------------.<++++++++.--------.++ +.------.--------.>+." > brainfxxk(hello) [1] "Hello, world!" http://www.slideshare.net/mickey24/rbrainfck-1085191
  • 5. Tsukuba.R#5 ? Animation with R ? library(animation) http://d.hatena.ne.jp/mickey24/20090614
  • 6. Tsukuba.R#6 ? Extend R with C!!! ? C R (C OpenCV ) http://d.hatena.ne.jp/mickey24/20091123/r_de_extension
  • 9. ? ?n snow ?
  • 10. n
  • 11. ? ? CPU !"#! !"#! !"#! !!!! !!!! !!!! !"#!
  • 12. 1CPU sapply(1:8, function(x){ x^2 }) (function(x){x^2})(1) (function(x){x^2})(2) (function(x){x^2})(3) (function(x){x^2})(4) !"#! (function(x){x^2})(5) (function(x){x^2})(6) (function(x){x^2})(7) (function(x){x^2})(8) [1] 1 4 9 16 25 36 49 64
  • 13. 4CPU sapply(1:8, function(x){ x^2 }) (function(x){x^2})(1) (function(x){x^2})(3) (function(x){x^2})(5) (function(x){x^2})(7) (function(x){x^2})(2) (function(x){x^2})(4) (function(x){x^2})(6) (function(x){x^2})(8) !"#! !"#! !"#! !"#! [1] 1 4 9 16 25 36 49 64
  • 14. ? ? ? ? 1CPU 1 100CPU 10
  • 15. R ? snow ? R ? http://cran.r-project.org/web/packages/snow/index.html ? R apply ? ? Socket PVM MPI
  • 17. snow ? 2CPU ? ? CPU
  • 18. ? snow > install.packages("snow") ? ?
  • 19. matprod.R ? 1000 n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) C <- A %*% B ?
  • 20. clmatprod.R library(snow) n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) cpu <- 2 hosts <- rep("localhost", cpu) cl <- makeCluster(hosts, type="SOCK") C <- parMM(cl, A, B) # C <- A %*% B stopCluster(cl)
  • 22. ? ? parMM(cl, A, B) ? apply ? parApply(cl, X, MARGIN, fun, ...) ? parLapply(cl, X, fun, ...) ? parSapply(cl, X, fun, ..., simplify=TRUE, USE.NAMES=TRUE) ? etc. ? ? clusterMap(cl, fun, ..., MoreArgs = NULL, RECYCLE = TRUE) ? clusterCall(cl, fun, ...) ? etc. ? help (?parApply, ?clusterMap )
  • 23. snow
  • 24. ? ? snow ? mCPU n ? O(n^3) #! $! !"#! !"#! !"#! !!!! !"#! #! !! "!
  • 25. ? 1CPU( ) system.time(A %*% B) ? mCPU ( ) system.time(parMM(cl, A, B))
  • 26. ? @DBCLS ? ? Sun Grid Engine OpenMPI ?
  • 27. @DBCLS 1 8 CPU 8CPU 64CPU 16GB 128GB !"#$! !"#$! !"#$! !"#$! !"#$! !"#$! !"#$! !"#$!
  • 29. 1000 ( ) 3.00 2.50 2.00 1.50 1.00 Faster! 0.50 0 1 4 8 16 CPU 4CPU 1.14
  • 30. 3000 ( ) 40.00 30.00 20.00 10.00 Faster! 0 1 4 8 16 CPU 8CPU 2.70
  • 32. ? snow ? C ? C R (40 50 ) www http://d.hatena.ne.jp/ syou6162/20090117/1232120983
  • 33. ? snow ? snow Simpli?ed http://www.sfu.ca/~sblay/R/snow.html ? RjpWiki - R http://www.okada.jp.org/RWiki/?R%A4%C7%CA%C2%CE%F3%B7%D7%BB%BB ? RjpWiki - L.Tierney snow http://www.okada.jp.org/RWiki/?L.%20Tierney%BB%E1%A4%CEsnow %A5%D1%A5%C3%A5%B1%A1%BC%A5%B8%A4%C7%A5%AF %A5%E9%A5%B9%A5%BF%B7%D7%BB%BB%A4%F2%B9%D4%A4%A6#scce80a1 ? Rmpi + snow + Sun Grid Engine ? Scheduled Parallel Computing with R: R + Rmpi + OpenMPI + Sun Grid Engine (SGE) http://blog.nguyenvq.com/2010/01/20/scheduled-parallel-computing-with-r-r-rmpi- openmpi-sun-grid-engine-sge/
  • 36. R+ Sun Grid Engine + Open MPI + snow
  • 37. ? @DBCLS ? Sun Grid Engine ? OpenMPI ? mpirun R ? Rmpi snow OpenMPI ? ?
  • 38. ? ? Rmpi OpenMPI ? snow Rmpi > install.packages("Rmpi", configure.args= "/path/to/mpidir") > install.packages("snow")
  • 39. gmatprod.R library(Rmpi) library(snow) n <- 1000 A <- matrix(rnorm(n^2), n) B <- matrix(rnorm(n^2), n) cl <- makeCluster() # C <- parMM(cl, A, B) # C <- A %*% B stopCluster(cl)
  • 40. gmatprod.sh #!/bin/bash #$ -S /bin/bash # /bin/bash #$ -j y # stdout stderr -o #$ -o gmatprod.log # stdout stderr #$ -pe openmpi 8 # export PATH=[path to R & snow_install_dir & mpirun]:$PATH export LD_LIBRARY_PATH=/path/to/mpidir/lib:$LD_LIBRARY_PATH MPIRUN_PATH=/path/to/mpirun MPIRUN_OPTS="-np ${NSLOTS} -machinefile ${TMPDIR}/machines" RSCRIPT=gmatprod.R RPATH=/path/to/snow_install_dir/RMPISNOW CMD="${RPATH} CMD BATCH --no-save ${RSCRIPT}" ${MPIRUN_PATH} ${MPIRUN_OPTS} ${CMD}
  • 41. ? qsub Grid Engine $ qsub gmatprod.sh
  • 42. ? ? Scheduled Parallel Computing with R: R + Rmpi + OpenMPI + Sun Grid Engine (SGE) http://blog.nguyenvq.com/2010/01/20/scheduled-parallel- computing-with-r-r-rmpi-openmpi-sun-grid-engine-sge/ ? Sun Grid Engine MPI R