Nick Felt wrote a script that you can use that preforms these set-up steps for you. You can copy the script from here, set it to executable, and run it:
cp ~newhall/public/cs87/xsede_setup.sh chmod 755 xsede_setup.sh ./xsede_setup.sh
Here is detailed directions of what to follow in Andy's pages and in what order:
You need xsede and longhorn accounts Follow steps of Andy's http://svn.cs.swarthmore.edu/gis_mpi/wiki/XsedeOverview (1) See the XsedeAccount page for details .. (a) follow the gsissh login steps (b) follow the XsedeSSH link for creating and installing ssh keys (c) then follow the rest of the ssh keys ssh-add stuff (d) then try: ssh tg-longhorn this will be the regular way to connect from now on (2) Trying out some stuff on longhorn (a) Configuring longhorn for gpu/mpi use (svn.cs.swarthmore.edu/gis_mpi/wiki/Longhorn/Build) (i) chsh password: /bin/bash (ii) add entries to .profile_user (iii) source .profile_user (3) Try out some of Andy's tg-demos (a) git clone user@cs.swarthmore.edu:/home/adanner/public/repos/tg-demos (http://svn.cs.swarthmore.edu/gis_mpi/wiki/XsedeOverview) (b) buid the tg-demos using cmake (4) Running on Longhorn link look at http://svn.cs.swarthmore.edu/gis_mpi/wiki/Longhorn/Tutorial (a) mpi-hello (i) cd jobscripts (ii) edit jobscripts/.sge_request and replace adanner email address with your address. mkdir ~/logs Make a logs folder in your home directory. qsub longhorn-mpi-hello qstat (b) longhorn user's guide (https://www.xsede.org/web/guest/tacc-longhorn) running batch jobs on Longhorn: qsub subfile qstat fields explained (search for "Batch query"): https://www.xsede.org/web/guest/tacc-lonestar#running qdel jobID
source files are here: /usr/local/CUDA_SDK/C/src/ the binaries are here: /usr/local/CUDA_SDK/C/bin/linux/release/
/usr/local/stow/cuda-4.0-sdk/CUDA_SDK/C/bin/linux/release/deviceQueryfrom this you can see the number of cores, the size of GPU memory, the max number of CUDA threads per block, max number of dimentions of blocks, the max threads per dimention, ... For this project, using a 2D 500x500 blocks, each with a single thread, is a size that is supported on all CS lab machines.
cp -r ~newhall/public/cs87/pthreads_example .
cp -r ~newhall/public/cs87/pthreads_signals_example .
cp -r ~newhall/public/cs87/socket_startingpt .
cp ~newhall/public/cs87/pthreads_example/* .
Create a cs87/gdb subdirectory, and copy over my gdb example programs into it:
/home/newhall/public/gdb_examplesCreate a cs87/valgrind subdirectory and copy over my valgrind example programs into it:
/home/newhall/public/purify_valgrind_examplesusing gdb to debug pthreads programs
(the above links are also available off my Unix and CS help pages)
cp -r ~newhall/public/cs87/lab1 .We will take a look at this code, which shows some examples of where C differs from C++, and an example of dividing a C program into multiple modules (.c files), using .h files, using static and extern definitions, and writing and using a makefile to build an executable.
Create a cs87/gdb subdirectory, and copy over my gdb example programs into it:
/home/newhall/public/gdb_examplesCreate a cs87/valgrind subdirectory and copy over my valgrind example programs into it:
/home/newhall/public/purify_valgrind_examples
(the above links are also available off my Unix and CS help pages)