Programming with Big Data in R
An editor has nominated this article for deletion. You are welcome to participate in the deletion discussion, which will decide whether or not to retain it. |
 | The topic of this article may not meet Wikipedia's general notability guideline. (June 2013) |
| This article needs attention from an expert in Computer science. Please add a reason or a talk parameter to this template to explain the issue with the article. (June 2013) |
| pbdR | |
|---|---|
| File:Pbdr.png | |
| Paradigm | SPMD |
| Designed by | Wei-Chen Chen, George Ostrouchov, Pragneshkumar Patel, and Drew Schmidt |
| Developer | pbdR Core Team |
| First appeared | Sep. 2012 |
| Preview release | |
| Typing discipline | Dynamic |
| OS | Cross-platform |
| License | General Public License and Mozilla Public License |
| Website | r-pbd.org |
| Influenced by | |
| R, C, Fortran, and MPI | |
Programming with Big Data in R (pbdR)[1][2][3] is a series of R packages and an environment for statistical computing with Big Data by utilizing high-performance statistical computation.[4] The pbdR uses the same programming language as R[5] with S3/S4 classes and methods which is used among statisticians and data miners for developing statistical software. The significant difference between pbdR and R[5] codes is pbdR mainly focuses on distributed memory system where data are distributed across several processors and analyzed in a batch mode, while communications between processors are based on MPI which is easily utilized in large high-performance computing (HPC) systems. R system[5] mainly focuses on single multi-core machines for data analysis via an interactive mode such as GUI interface.[6]
Two main implementations in R using MPI are Rmpi[7] and pbdMPI of pbdR.
- The pbdR built on pbdMPI uses SPMD parallelism where every processors are considered as workers and own parts of data. The SPMD parallelism[8][9] introduced in mid 1980 is particularly efficient in homogeneous computing environments for large data, for example, performing singular value decomposition[10] on a large matrix, or performing clustering analysis on high-dimensional large data. On the other hand, there is no restriction to use manager/workers parallelism in SPMD parallelism environment.
- The Rmpi[7] uses manager/workers parallelism where one main processor (manager) servers as the control of all other processors (workers). The manager/workers parallelism[11] introduced around early 2000 is particularly efficient for large tasks in small clusters, for example, bootstrap method and Monte Carlo simulation in applied statistics since i.i.d. assumption is commonly used in most statistical analysis. In particular, task pull parallelism has better performance for Rmpi in heterogeneous computing environments.
The idea of SPMD parallelism is to let every processors do the same works but on different parts of large data. For example, modern GPU is a large collection of slower co-processors which can simply apply the same computation on different parts of relatively smaller data, but the SPMD parallelism ends up an efficient way to obtain final solutions, i.e. time to solution is shorter.[12][13] It is clearly that pbdR is not only suitable for small clusters, but also is stabler for analyzing Big data and is more scalable for supercomputers.[14] In short, pbdR
- does not like Rmpi, snow, snowfall, do-like, nor parallel packages in R,
- does not focus on interactive computing nor master/workers,
- but is able to use both SPMD and task parallelisms.
Package design
Programming with pbdR requires usage of various packages developed by pbdR core team. Packages developed are the following.
| General | I/O | Computation | Application |
|---|---|---|---|
| pbdDEMO | pbdNCDF4 | pbdDMAT | pmclust |
| pbdMPI | pbdBASE | ||
| pbdSLAP |
Among these packages, pbdMPI provides wrapper functions to MPI library, and it also produces a shared library and a configuration file for MPI environments. All other packages rely on this configuration for installation and library loading that avoid difficulty of library linking and compiling. All other packages can directly utilize MPI functions easily.
- pbdMPI --- an efficient interface to MPI either OpenMPI[15] or MPICH2[16] with a focus on Single Program/Multiple Data (SPMD) parallel programming style[8][17][9]
- pbdSLAP --- bundles scalable dense linear algebra libraries in double precision for R, based on ScaLAPACK version 2.0.2[18] which includes several scalable linear algebra packages (namely BLACS, PBLAS, and ScaLAPACK).[19][20]
- pbdNCDF4 --- Interface to Parallel Unidata NetCDF4 format data files[21]
- pbdBASE --- low-level ScaLAPACK codes and wrappers
- pbdDMAT --- distributed matrix classes and computational methods, with a focus on linear algebra and statistics[22][23][24]
- pbdDEMO --- set of package demonstrations and examples, and this unifying vignette[3]
- pmclust -- parallel model-based clustering using pbdR
Amount those packages the pbdDEMO package is a collection of 20+ package demos which offer example uses of the various pbdR packages, and contains a vignette which offers detailed explanations for the demos and provides some mathematical or statistical insight.
Examples
Example 1
Hello World! Save the following code in a file called ``demo.r``
### Initial MPI
library(pbdMPI, quiet = TRUE)
init()
comm.cat("Hello World!\n")
### Finish
finalize()
and use the command
mpiexec -np 2 Rscript demo.r
to execute the code where Rscript is one of command line executable program.
Example 2
The following example modified from pbdMPI illustrates the basic syntax of the language of pbdR. Since pbdR is designed in SPMD, all the R scripts are stored in files and executed from the command line via mpiexec, mpirun, etc. Save the following code in a file called ``demo.r``
### Initial MPI
library(pbdMPI, quiet = TRUE)
init()
.comm.size <- comm.size()
.comm.rank <- comm.rank()
### Set a vector x on all processors with different values
N <- 5
x <- (1:N) + N * .comm.rank
### All reduce x using summation operation
y <- allreduce(as.integer(x), op = "sum")
comm.print(y)
y <- allreduce(as.double(x), op = "sum")
comm.print(y)
### Finish
finalize()
and use the command
mpiexec -np 4 Rscript demo.r
to execute the code where Rscript is one of command line executable program.
Example 3
The following example modified from pbdDEMO illustrates the basic ddmatrix computation of pbdR which performs singular value decomposition on a given matrix. Save the following code in a file called ``demo.r``
# Initialize process grid
library(pbdDMAT, quiet=T)
if(comm.size() != 2)
comm.stop("Exactly 2 processors are required for this demo.")
init.grid()
# Setup for the remainder
comm.set.seed(diff=TRUE)
M <- N <- 16
BL <- 2 # blocking --- passing single value BL assumes BLxBL blocking
dA <- ddmatrix("rnorm", nrow=M, ncol=N, mean=100, sd=10)
# LA SVD
svd1 <- La.svd(dA)
comm.print(svd1$d)
# Finish
finalize()
and use the command
mpiexec -np 2 Rscript demo.r
to execute the code where Rscript is one of command line executable program.
Further reading
- Introduction to distributed computing with pbdR at the UMBC High Performance Computing Facility (Technical Report, 2013).[25]
- CRAN Task View: High-Performance and Parallel Computing with R.[26]
- R at 12,000 Cores.[27] This article was read 22,584 times in 2012 since it posted on October 16, 2012 and ranked number 3 according to 100 R posts of 2012[28]
- Profiler for pbdR mentored by the Organization of R Project for Statistical Computing for Google summer of code 2013.[29]
External links
- Official website of the pbdR project
- Technical website of the pbdR packages
- Source Code of developing version of the pbdR packages
- Discussion Group for any of pbdR related topics
Milestones
2013
- Version 1.0-2: Add pmclust.
- Version 1.0-1: Add pbdNCDF4.
- Version 1.0-0: Add pbdDEMO.
2012
- Version 0.1-2: Add pbdBASE and pbdDMAT.
- Version 0.1-1: Add pbdSLAP.
- Version 0.1-0: Migrate from Rmpi[7] to pbdMPI.
References
- ^ Ostrouchov, G., Chen, W.-C., Schmidt, D., Patel, P. (2012). "Programming with Big Data in R". URL http://r-pbd.org.
{{cite journal}}: External link in|journal= - ^ "XSEDE".
- ^ a b Schmidt, D., Chen, W.-C., Patel, P., Ostrouchov, G. (2013). "Speaking Serial R with a Parallel Accent" (PDF).
{{cite journal}}: Cite journal requires|journal=(help)CS1 maint: multiple names: authors list (link) - ^ Chen, W.-C. and Ostrouchov, G. (2011). "HPSC -- High Performance Statistical Computing for Data Intensive Research".
{{cite web}}: CS1 maint: multiple names: authors list (link) - ^ a b c R Core Team (2012). R: A Language and Environment for Statistical Computing. ISBN 3-900051-07-0.
- ^ Martinez, W. L. (2011), Graphical user interfaces. WIREs Comp Stat, 3: 119–133. doi: 10.1002/wics.150
- ^ a b c Yu, H. (2002). "Rmpi: Parallel Statistical Computing in R". R News.
- ^ a b Darema, F. (2001). "The SPMD Model: Past, Present and Future".
{{cite journal}}: Cite journal requires|journal=(help) - ^ a b Ostrouchov, G. (1987). "Parallel Computing on a Hypercube: An Overview of the Architecture and Some Applications". Proc. 19th Symp. on the Interface of Computer Science and Statistics: 27-32.
- ^ Golub, Gene H.; Van Loan, Charles F. (1996). Matrix Computations (3rd ed.). Johns Hopkins. ISBN 978-0-8018-5414-9.
- ^ "Google's MapReduce Programming Model -- Revisited" — paper by Ralf Lämmel; from Microsoft
- ^ Darren Murph. "Stanford University tailors Folding@home to GPUs". Retrieved 2007-10-04.
- ^ Mike Houston. "Folding@Home - GPGPU". Retrieved 2007-10-04.
- ^ Schmidt, D., Ostrouchov, G., Chen, W.-C., and Patel, P. (2012). "Tight Coupling of R and Distributed Linear Algebra for High-Level Programming with Big Data". High Performance Computing, Networking, Storage and Analysis (SCC), 2012 SC Companion:: 811–815.
{{cite journal}}: CS1 maint: extra punctuation (link) CS1 maint: multiple names: authors list (link) - ^ Jeff Squyres. "Open MPI: 10^15 Flops Can't Be Wrong" (PDF). Open MPI Project. Retrieved 2011-09-27.
- ^ MPICH License
- ^ Ortega, J.M., Voight, G.G., and Romine, C.H. (1989). "Bibliography on Parallel and Vector Numerical Algorithms".
{{cite journal}}: Cite journal requires|journal=(help)CS1 maint: multiple names: authors list (link) - ^ Blackford, L.S.; et al. (1997). ScaLAPACK Users' Guide.
{{cite book}}: Explicit use of et al. in:|author=(help) - ^ Petitet, Antoine (1995). "PBLAS". Netlib. Retrieved 13 July 2012.
- ^ Jaeyoung Choi; Dongarra, J.J.; Walker, D.W. (1994). "PB-BLAS: a set of Parallel Block Basic Linear Algebra Subprograms" (PDF). Scalable High-Performance Computing Conference: 534–541. doi:10.1109/SHPCC.1994.296688. ISBN 0-8186-5680-8.
{{cite journal}}: Unknown parameter|month=ignored (help) - ^ NetCDF Group (2008). "Network Common Data Form".
- ^ J. Dongarra and D. Walker. "The Design of Linear Algebra Libraries for High Performance Computers".
{{cite journal}}: Cite journal requires|journal=(help) - ^ J. Demmel, M. Heath, and H. van der Vorst. "Parallel Numerical Linear Algebra".
{{cite journal}}: Cite journal requires|journal=(help)CS1 maint: multiple names: authors list (link) - ^ "2d block-cyclic data layout".
- ^ Raim, A.M. (2013). Introduction to distributed computing with pbdR at the UMBC High Performance Computing Facility (Technical report). UMBC High Performance Computing Facility, University of Maryland, Baltimore County. HPCF-2013-2.
- ^ Dirk Eddelbuettel. "High-Performance and Parallel Computing with R".
- ^ "R at 12,000 Cores".
- ^ "100 most read R posts in 2012 (stats from R-bloggers) – big data, visualization, data manipulation, and other languages".
- ^ GSOC-R 2013. "Profiling Tools for Parallel Computing with R".
{{cite web}}: CS1 maint: numeric names: authors list (link)