本机 ssh-keygen -t rsa – cd ~/ssh – cp -p id_rsa.pub authorized_keys2 – chmod go-rwx authorized_keys2 – ssh-agent $SHELL; ssh-add ssh-copy-id -i id_rsa.pub uwhpsc@166.111.138.172
http://unixlab.sfsu.edu/~whsu/EXCL/656proj/tutorial
open-mpi没有p4pg
Getting ready for Project 5 --- MPI tutorial -------------------------------------------- For Project 5, you will be writing a simple parallel program to run on a network of workstations in Science 252. This handout contains preliminary information for you to get setup and run some examples. SETTING UP THE MACHINES MPI stands for Message Passing Interface. It is becoming the standard for parallel programming. You will be running programs using the MPI library on a network of workstations in Sci 252. The network of PCs in Sci 252 can boot either Windows NT or Sun Solaris. You must make sure that the machines you intend to work on are running Solaris. The MPI installation comes with a "tstmachines" script that should test if machines are available; unfortunately we have not been able to get this to work. If you can successfully "finger" a PC in Sci 252, it should be running Solaris and available to you for running MPI programs. The server for the network, sci-252pc20, always runs Solaris. If you are in the lab and need to reboot an NT station to run Solaris, go to the initial NT window and click on "Shutdown". Make sure the option "Shutdown and restart" is selected, and allow the machine to proceed. When you come to a screen for selecting NT and Solaris, pick Solaris (of course). Everything else should be default; when in doubt just hit ENTER. For initial development, I suggest using a small number of machines (like 2 or 3). Get that to work, then move to a larger group. If you wrote clean code, there shouldn't be major problems. The common mpi commands are in /usr/local/mpi/bin. You should add this to your path. SETTING UP SSH (SECURE SHELL) When you log on for the first time, you will need to setup ssh. (Setting it up on one of the clients will get you setup on all of them.) Follow this procedure: prompt> /usr/local/bin/ssh-keygen prompt> cp ~/.ssh/identity.pub ~/.ssh/authorized_keys prompt> chmod 600 ~/.ssh/authorized_keys prompt> chmod 700 ~/.ssh prompt> chmod 644 ~/.ssh/identity.pub prompt> /usr/local/bin/ssh-agent $SHELL prompt> /usr/local/bin/ssh-add When you run ssh-keygen, you will be prompted for a passphrase. This is like a password; it's case-sensitive and can include white spaces. Think of a phrase you can remember easily, and DO *NOT* SAVE IT IN A FILE!!!! When you run ssh-add, you will be prompted for your passphrase. You have to type all 7 lines the first time you log in. After that, every time you log in, you only have to type the last two lines to setup ssh and run MPI programs. RUNNING SOME EXAMPLE PROGRAMS Once you're set up, copy the example programs cpi.c and sr.c from my directory ~hsu/mpi_ex on sci-252pc20 (or any of the other machines in Sci 252). sr.c only involves two processors. Processor 0 sends a message to processor 1. Processor 1 waits for this message, and sends a message back to processor 0. Processor 0 measures the roundtrip time for a message, for 20 send-receive pairs. cpi.c is a somewhat cleaner version of the pi computation program that you saw in the book chapter copies you received in class. Processor 0 reads the value of n from the user, and broadcasts it to all processors. Each processor does its computations, all partial results are accumulated using MPI_Reduce(), and processor 0 prints the result and the execution time. The times reported (using MPI_Wtime() calls) is in 1000s of seconds (when the display says 0.1, it means 100 seconds). To compile an MPI program (for example, cpi.c), type mpicc cpi.c -o cpi To run an MPI program, log onto sci-252pc14, and type mpirun -p4pg pf4 cpi (pf4 is a file that specifies how processes are assigned to hosts. More on that later.) You'll get output similar to this (don't worry about the stty error messages for now): stty: : No such device or address stty: : No such device or address stty: : No such device or address Process 0 on sci-252pc14.sci252_cs.sfsu Process 1 on sci-252pc20.sci252_cs.sfsu Process 2 on sci-252pc6.sci252_cs.sfsu Process 3 on sci-252pc7.sci252_cs.sfsu Enter the number of intervals: (0 quits) 10000 [user entered 10000] pi is approximately 3.1415926544231239, Error is 0.0000000008333307 wall clock time = 0.000000 Enter the number of intervals: (0 quits) 40000 [user entered 40000] pi is approximately 3.1415926536418795, Error is 0.0000000000520863 wall clock time = 0.000002 Enter the number of intervals: (0 quits) 0 [user entered 0] P4 procgroup file is pf4. prompt> Your output may also be scrambled a little bit, since you have several processes printing to the same screen, and order was not enforced. As we mentioned earlier pf4 gives mpirun directions on how to assign processes to processors. The pf4 included in your examples looks like this: sci-252pc14 0 /Users/hsu/mpi_ex/cpi sci-252pc20 1 /Users/hsu/mpi_ex/cpi sci-252pc6 1 /Users/hsu/mpi_ex/cpi sci-252pc7 1 /Users/hsu/mpi_ex/cpi Each line is a triplet: [hostname] [#processes] [program name]. The first line indicates that the job was started on pc14; notice that the number of processes specified for the host that started the job is always 0. The next three lines indicate that one process is to be started on pc20, pc6 and pc7. They will run the program /Users/hsu/mpi_ex/cpi. You can also put multiple processes on the same host. For example, to put two processes on pc6 and start it from pc6, you would make a file: sci-252pc6 0 /Users/hsu/mpi_ex/cpi sci-252pc7 1 /Users/hsu/mpi_ex/cpi
