Next 17: Graphical Tools  Up 16: Programming  Prev Compiling Fortran  Prev Compiling C  Contents

Specialized

§ 16.C: Programming Courtesy


When running programs on a shared machine, it is important to be aware of how your program consumes resources. If your program consumes many resources, you must be careful not to interfere with other users. If you are working with large arrays, or if your program takes more than a few minutes (i.e. about 1/2 hour or more) you can reasonably assume that it is affecting other users.

If you need a lot of memory (greater than 32 Meg) you will want to use goophy or daphy to do your work. However, goophy and daphy are primarily for class use. Therefore, if you need to run a large research program on the cluster, it is recommended that you do so between 10 pm and 8 am. If you run a large job on goophy or daphy during the daytime, it will be killed.

If your program takes a long time to run, but not a lot of memory you should run it on one of the fruit loop (banana, coconut, guava, graphy, mango or papaya) machines. In addition, out of consideration for other users, you should use nice. Nice places your job at a lower priority (it can also be used to place your job at a higher priority, but you should not have permissions for that) than it would otherwise be at, allowing other programs to run ahead of it. To use nice type nice followed by -15 (the number fifteen is a parameter you can read about in the man pages if you are curious) followed by the command you want niceed. For example, the to run the program a.out with nice:
> nice -15 a.out

While nice can help keep your job from interfering with other users, it is not a guarantee against such interference. Because class use has priority over research on the cluster machines, if your job is making it difficult for other users to use the cluster machines, it will be terminated. If you have a very long or memory intensive job, it is suggested that you run it at night, or at some other low usage time.

To get control of your terminal window while running a long program, you should run it in the background. To do this type the & after your command. For instance:
> nice -15 a.out &

You can now log out without stopping your program.

If you decide you want control back after you have started your program you can use the ^ Z (ctrl-Z) command to put your program in the background.

If you start a program without nice you can lower its priority with renice. To renice something you need the job number. To get the job number use the ps (Process Status) command. If you try a ps
> ps

you might get something like


  PID TTY      S           TIME COMMAND
11093 ttyp1    S        0:01.40 -tcsh (tcsh)
13805 ttyp1    R        0:00.02 a.out

The far right column lists the commands running under this session. The -tcsh line is your terminal and it, or something similar will come up whenever you use the ps command. Sometimes the ps command itself comes up, sometimes it does not. Once you have found the command you want to renice look in the far left column for the process identification number (PID). In the example given a.out has the process number 13805. To renice this command you would type.
> renice -15 13805

If you should decide you no longer want to run a job (for instance if it is taking too long), you can stop it. If it is running in the foreground (your terminal is locked up waiting for it) you can use the ^ c (ctrl-c) command to stop it.

If it is running in the background you can use the you can use the kill command. The kill command also needs the job number. To kill the example program above you would type:
> kill -9 13805

A kill may not always work on the first try. You should use a ps after a few minutes to check that the program is no longer running. If the program is still running re-enter your kill command.

If you remain aware of how you are modifying a program you should be able to estimate the time and memory requirements of your as your programs become larger. Be careful to consider the multiplicative properties of multidimensional arrays. If you double the number of each dimension in your 3-D array, you can expect your memory needs to increase by a factor of eight. Your program time will also likely increase by a factor of eight or more. If you are careful and considerate about when and how you use the cluster resources, we can keep the cluster working well for all users.


Next 17: Graphical Tools  Up 16: Programming  Prev Compiling Fortran  Prev Compiling C  Contents

Comments and questions to nacse-unix@physics.orst.edu.