Running on Supermike -------------------- SuperMike has 4 head nodes and 512 compute nodes. Hostnames for the head nodes are mike1.lsu.edu mike2.lsu.edu mike3.lsu.edu and mike4.lsu.edu The head nodes are identical. Access to the SuperMike cluster is via secure shell client (ssh) to one of the head nodes. 1. To allow communications among the nodes type: ------------ cp -f /etc/skel/.ssh/known_hosts ~/.ssh/known_hosts 2. Compilers available include: ------------------------------- gcc,g77 ,g++ (GNU) pgcc,pgCC, pgf77,pgf90 (portland Group) icc, ifc (Intel) We, CACTUS users want to use intel. 3. MPI ------ There are three versions of MPICH compiled for each of the compilers installed , GNU, Portland Group and Intel Depending on the version of MPICH you are going to use, ionly one of the these three lines should be uncommented out in your .bashrc file: #export MPI_HOME=/usr/local/gmpi_pgi #export MPI_HOME=/usr/local/gmpi_gcc #export MPI_HOME=/usr/local/gmpi_intel mpicc mpif77 mpif90 are the commands to compile your mpi programs. We CACTUS users want to uncomment the third line. 4. PBSPro is our queuing system. -------------------------------- Submitting a job to the PBS queue requires a PBS script. A working script example using 32 nodes, two nodes per processor is: #!/bin/bash #PBS -m b #PBS -m e #PBS -q c32 #PBS -l nodes=32:ppn=2 #PBS -l walltime=24:00:00 #PBS -o #PBS -j oe #PBS -N #PBS #cd $PBS_O_WORKDIR # date cd # a MPI 64 processors run mpirun.ch_gm -machinefile $PBS_NODEFILE -np 64 # date you just need to fill with your need the names in < >'s. 5. Many queues are configured: ------------------------------ Queue Memory CPU Time Walltime Node Run Que Lm State ---------------- ------ -------- -------- ---- ----- ----- ---- ----- reserved_big -- -- -- 256 0 0 1 E R workq -- -- 48:00:00 8 0 0 4 E R c128_1 -- -- 24:00:00 128 1 3 1 E R c64 -- -- 40:00:00 64 2 5 5 E R c32 -- -- 24:00:00 32 4 0 6 E R c16 -- -- 24:00:00 16 10 3 10 E R fast -- -- 01:00:00 2 0 0 -- E R glabtest -- -- -- 2 0 0 3 E R c64_tmp -- -- 80:00:00 64 0 0 -- E R ----- ----- 17 11 6. Running You have to type in the command line: qsub -q ./your-pbs-script e.g qsub -q c16 ./your_pbs_script will launch your job 6. interactive runs ------------------- you can use the fast queue for interactive runs: a. qsub -q fast -I b. gmpiconf > machines c. mpirun.ch_gm -machinefile machines -np 2 ./executable parfile 7. COMPILING CACTUS ------------------ An options file would be the following, you can also use the cactus cpp: CPP=/home1/tradke/bin/cpp CC=icc CXX=icc F90=ifc HDF5=yes HDF5_DIR=/exports/local/packages/hdf5-1.6.1-intel/serial LIBSZ_DIR=/exports/local/packages/hdf5-1.6.1-intel/serial # cannot use MPI=MPICH because it needs another external library # change MPI path according to the compilers used MPI=CUSTOM MPI_LIB_DIRS=/usr/local/gmpi_intel/lib /usr/gm/lib MPI_INC_DIRS=/usr/local/gmpi_intel/include MPI_LIBS=mpich gm