Computational Methods for Statistic with Applications Computer
Contents
1. Lower the computational load Consider y lt function x z lt x73 Do you have improvement in the run time over Lapply when you use mclapply Why 5 Choose your own function and experiment Have a look at the user manual of multicore there are more possibilities to explore 6 Since Kalkyl has also multicore processors you could repeat your experiments there Proceed as follows ssh AX kalkyl uppmax uu se interactive A g2011131 t 0 15 00 module unload pgi module load gcc R gt install packages multicore library multicore Again via top you could see 8 threads running in parallel Utilizing Kalkyl as a cluster Exercise 2 R packages Rmpi snow Experience the packages Rmpi and snow and how to submit jobs to the queueing system Tacks 1 You should read have read the user guide http www uppmax uu se support user guides kalkyl user guide It is expected that you understand how the batch scripts look like how to check the status of the submitted jobs cancel submitted or running jobs etc 2 Install the necessary packages as follows module unload pgi openmpi module load gcc openmpi R gt install packages Rmpi gt install packages snow 3 Repeat the example from the lecture notes to compute a double integral in parallel 2 2 Iie 3x y 3y dz dy 4 5 SESI To this end you are given the thr2. Computational Methods for Statistic with Applications Computer Exercise no 5 Sparse matrices NGSSC LU SLU UU September 2011 The goal with this lab is that you get acquainted with some packages in R which enable parallel computations At the end selected results of the exercises have to be sketched and sent to the lab consultant Some examples of functions and script files can be downloaded via http user it uu se maya Courses NGSSC index_Stat html Make a copy of these files in some of your directories Utilizing a multicore processor Exercise 1 R package multicore Tasks 1 Login on some of the multicore Unix machines linne it uu se scheele it uu se sernander it uu se svedberg it uu se tiselius it uu se fries it uu se polhem it uu se These are 8 core machines and we can use Shared Memory type of parallelism to execute com putation in parallel Please coordinate with your colleagues so that not too many of you login on one and the same machine 2 Start R and install the package multicore 3 Perform the following test and check the time spent for various values of n 10 10 10 107 library multicore y lt function x z lt x 3 sqrt x x 1 3 x lin system time lapply x y system time mclapply x y Check if you are running in parallel in a separate terminal window type top and you should see multiple tasks with your user name running simultaneously 4
3. ee R routines which compute the integral in a serial fashion integLoop integVec integApply a It is recommended to check the performance of the three versions to also see the effect of utilizing the vectorization in the computations Check and run integr_serial_run r b For the parallel experiments you have at your disposal slavefunc integRmpi integSnow integr_Rmpi_run r integr_Snow_run r and examples of batch files script_run_Rmpi_integr script_run_Snow_integr 4 You are given another simple example for computing the row sum of a matrix snow_rowsum r and script_run_snow_rowsum It is suggested that you run the example monitor the par allel performance and reason on the timing results is it scalable Use different sizes and number of processes Prepare a summary of the timing results for various number of processes and problem sizes Exercise 3 More packages Install the packages foreach and doMC Then try require foreach require doMC search n 1000 loads pa loads bo make sur size of registerDoMC cores 2 tell x lt foreach i 1 10 dopar ckage foreach th doMC and multicore e all 3 packages are loaded the matrices s R to spawn a maximum of 2 processes svd matrix rnorm n xn ncol n Change the number of threads test with various sizes What is the speedup you obtain
Download Pdf Manuals
Related Search