PSU CS Department
Experimental Safety Guidelines

It sounds ludicrous to be talking about safety in connection with computing experiments, right? After all, what could happen? No experiment you are likely to run will have any interesting effect whatsoever on the physical world.

Of course, there's a social world. What you want to avoid is running an important department computer out of CPU overnight, and thus getting thwacked senseless by your own classmates. This is what we mean by ``experimental safety.''

Thus, please follow these helpful suggestions when conducting compute-intensive experiments:

• Please stay off the department servers. Here's a list of machines that you may run your experiments on.

  antares.cs.pdx.edu
  betelgeuse.cs.pdx.edu
  fomalhaut.cs.pdx.edu
  alpheratz.cs.pdx.edu
  regulus.cs.pdx.edu
  castor.cs.pdx.edu
  pollux.cs.pdx.edu
  spica.cs.pdx.edu
  arcturus.cs.pdx.edu
  

  These machines are SPARC Ultra 5s in the lab. You may not run your experiments on rigel.cs.pdx.edu or sirius.cs.pdx.edu. For anything not in either of these categories, ask someone.

  If possible, be seated at the console when running experiments, in order to avoid surprise for the console user.

• Ask the operating system to limit your resource usage. Make the ulimit command your friend. In particular, ulimit -t 900 will set a CPU limit of 900 seconds (15 minutes) for all of your subsequent runs. Please do this, as no assignment I give you should need longer than this, and it will help avoid accidents.
• Similarly, ask the operating system to run your jobs at lower priority. The nice command does this: nice -15 foo runs foo at a very low priority. The renice command will do this for a running job: see the manual page.
• If someone asks you to cut it out, please do so: many people are trying to share a scarce resource, and this can be tough. Defer to the systems group on any and all issues related to resource usage.
• If you simply must have a beefy CPU for your trick algorithm, ask the instructor. They can almost always find you one.

Remember the golden rule of combinatorial search: better algorithms help, more CPU does not. Ten times the CPU gets you one or two nodes deeper in the tree: this kind of gain is not worth making everyone's life miserable, including your own.