computer guy
MIT Department of Mathematics Computer Help

Home
New User Help
Systems Available
Printing
Programs Available
Dial-up Access
Policies
FAQ
Problems
Valid XHTML 1.0!

Setting the priority of your compute jobs

This page lays out the guidelines in regards to running compute jobs that are cpu and/or memory intensive.

We currently have 4 compute hosts, one running Linux, the others Solaris. The Solaris machines archimedes and pythagoras have 2 CPUs and 1GB of memory, schauder, the other Solaris host has 2 slighly faster CPUs and 2GB of memory. The newest compute host is newton which 2 Athlon processors and 2GB of memory running Redhat Linux. We will be adding another high capacity Linux compute host in the coming months. These are our main hosts for any "heavy" computing, either via packages like matlab or maple or else by running long custom-made jobs.

A key point when using those machines is that whenever you submit a compute job (including matlab, mathematica, splus or maple) that is going to run in the background for more than about 2 hours, you should politely assign it the LOWEST priority possible. You can do this easily by prefixing "nice +19" at the time of submission, as in:

archimedes~::> nice +19 myspecialjob &

or you can subsequently lower the priority of one of your jobs that is already running via the command: "renice -n 19 -p 12345" if its PID = program ID happens to be 12345.

This way those who need a short interactive session with matlab or mathematica or some program of their own devising can do so quickly, without your background job getting too much in their way.

At this time, when someone attempts to run matlab, mathematica or matlab on an x-terminal server (lagrange, laguerre, lebesgue, laurent, kronecker, tarski or turing) a script is run informing the user to run the program on one of the compute hosts listed above. When the user thus starts the program on a compute host by default it is set to run at a moderate priority (nice +10).

There is one serious problem with matlab: If you logout or are logged out without exiting the program, it sometimes continues running, often at a high priority. So once you are done, check with 'ps auxw | grep matlab' for a runaway process. If you find it still running, please kill it at once.