MPI_2005.TEX

In October 2005, I presented several introductory talks on MPI, the Message Passing Interface, in order to acquaint our users with the the MPI library, the various local computing clusters that supported MPI, and the job control necessary to run MPI programs.

This was partly an attempt at publicity, but also an effort to encourage our graduate students to look further into MPI, as well as to prepare the groundwork for a talk on PETSC, to be given by our visitor Jeff Borggaard.

The LaTeX file used to create the presentation employs the Beamer class.

You can copy the source code for the slides:

• mpi_intro.tex, the LaTeX source;
• ../../pdf/mpi_intro.pdf, a PDF version;

The file includes various graphics files:

• computers.png, an image of an early "computer" room at Harvard College Observatory; the "computers" are women, doing various calculations based on data from astronomical observations.
• cray_ymp.png, an image of the Cray Y-MP supercomputer, a machine that depended largely on vectorization and pipelining. High speed required that the parts of the machine be very tightly clustered. This meant a great deal of concentrated heat, and so a major part of developing the Cray computers involved imaginative solutions to the problem of pumping coolant fast enough. The "benches" around the base of the machine and the odd "aquarium" in the foreground are actually parts of the cooling aparatus.
• eniac.png, the ENIAC computer at Princeton, created in 1942, and setting a record breaking speed of 5,000 additions in a second. The machine was used to compute ballistics tables, but John von Neumann insisted that it should be able to be used to predict the weather. In 1950, he published work suggestion how this should be done.
• europe_weather.png, a schematic of the computational grid employed by Lewis Richardson in his attempt to "predict" the weather using data from 25 sample points in Europe. The task of computing the weather 6 hours later, for just two of the points in the grid, required six months of work, and the results were disappointingly nonsensical.
• factory.png, after the failure of his first weather calculation, Lewis Richardson dreamed of a forecast factory or computational orchestra of 64,000 players, seated in a spherical hall representing the earth. Each player would have a slide rule or adding machine, working on the predicted weather at their local spot, and occasionally exchanging results with their neighbors. A conductor in the center of the hall would maintain order, using a bright light to point out players who were going too fast or too slow for the others.
• fsu_logo_ybkgrd.pdf, the FSU logo, wordmark, trade-doohickey, which is secretly branded on the inner thigh of every graduate.
• harvard_mark.png, the Harvard MARK computer, 50 feet long, 8 feet high, two feet deep. A single addition, subtraction, transfer or clear of data required 3/10 of a second. Multiplication took at most 6 seconds, division 16, while the computation of a logarithm, exponential or sine took 90, 66 or 60 seconds respectively. There was a staff of 10, including 4 mathematicians, 4 operators, and 2 maintenance men.
• hello_c.txt, the text of a C program, using MPI, to say HELLO 4 times.
• hello_cpp.txt, the text of a C++ program, using MPI, to say HELLO 4 times.
• hello_f77.txt, the text of a FORTRAN77 program, using MPI, to say HELLO 4 times.
• matvec_f77_page1.txt, piece 1 of a FORTRAN77 program, in which the "master" process sends out data to the workers.
• matvec_f77_page2.txt, piece 2 of a FORTRAN77 program, in which the "master" process receives results, and sends out more work.
• matvec_f77_page3.txt, piece 3 of a FORTRAN77 program, in which the workers receive data from the master, compute results, and send them back.
• nano.png, a picture of Grace Hopper, holding one of her "nanoseconds". This is a bit of cabling whose length is equal to the distance that light (or an electric signal) can travel in a nanosecond. She was forcefully making the point that the size of a computer with a clockspeed of one gigahertz could be no bigger than this length.
• wilma.png, a weather map showing the predicted path of hurricane Wilma. Although we snipe at the uncertainties of these predictions, we think nothing of the fact that a reasonable reliable forecast of tomorrow's weather can be computed and ready ...before tomorrow!