CNS Computing FAQ

Printing

How can I print to the ACS printers?

First of all, these printers work with Postscript, PDF, or plain text. If you have some other file, such as a LaTeX or Openoffice.org file, you must first generate a Postscript file. You can usually do this by saying “Print to File” or something of the sort. UNIX computers will then always generate Postscript files; under MS-Windows or Macintosh, you must select a Postscript printer driver to generate Postscript files. So, assuming you have a file printable by ACS:

Simplest solution: you can use scp to copy a file to acs.bu.edu, then ssh to that machine, and run lpr on that filename.

You can do it all in one line. To print foo.ps, type the following:

cat foo.ps | ssh acs "lpr"

There is a script which will do this for you; simply run acslpr [file1] [file2] .... Ensure you first set up SSH keys to acs, or it will repeatedly ask for your password.

Software Issues

How do I get started with LaTeX?

A fairly simple sample file demonstrates the most important features; it looks like this.

How do I format my thesis with LaTeX?

There are rather specific requirements. You might start with a template made by someone else.

I have to work with MS-Word.

That’s not a question, but may match your misfortune. You can translate your LaTeX code to XML using htlatex. The resulting file you can open in Openoffice.org, which can save in MS-Word format.

How can I generate monochrome patterned graphs in Matlab without resorting to bitmaps?

Apparently, you can’t. Instead, use Gnuplot, possibly along with Octave. For example, to generate a bar chart that looks like this PDF, you can use this simple Octave script, which requires Gnuplot 4 or newer. Usually you can simply save your data from Matlab, then load it into Octave and plot it from there. Octave uses Gnuplot to plot; the above script is needed since Octave does not support all the features of Gnuplot, and Gnuplot does not appear to have a built-in style for plots of this style.

Supercomputing

How do I use multiple CPUs in Matlab?

The simplest thing to do is to run two independent copies of Matlab, each of which will be about 75% as fast as if it were running on a separate computer. This allows you to get more work done in the same time, but not the same amount of work in less time.

But I want to speed up a single task.

I suggest you use Octave, the Free implementation of Matlab. There is a parallel version which you can dowload here.

Or you could use MPI with Octave, if it’s appropriate for your task.

More options are discussed here.

I depend on functions available only in proprietary Matlab.

You could rewrite to not depend on these functions. Alternatively, Rushi tells us that “the following helps a little, depending on what you are doing. Get the Intel MKL libraries, and set the following:

export OMP_NUM_THREADS=2
export LD_PRELOAD=/opt/intel/mkl61/lib/32/libguide.so
export BLAS_VERSION=/opt/intel/mkl61/lib/32/libmkl_p4.so

I haven’t followed this stuff, but the above was good as of Matlab 6.5. Of course, we are assuming Intel processors here. Similar solutions might exist with atlas (which is bundled with matlab beginning 7.0, I believe), or install your own atlas libraries and link to them in the above manner. For me the speedup was about 20% with the above.”


Valid HTML 4.01! Valid CSS! ©2005-2009 CaREMOVE THISmeron@MorAND THISland.ca Creative Commons License (2006-05-25 18:19)