Managing CPU resources

Contention for CPU resources is inevitable unless you run only one job. The following are recommendations and procedures you can use to ease that contention.

Balancing system work loads

Sometimes users do not realize the strain they are putting on system resources. Detecting problems in the workload and notifying users of alternatives can result in better system availability and response for everyone.

First, try to reduce the prime-time load.

• Work with system users to identify jobs that can be scheduled in other than prime-time.

• Schedule larger jobs for off-hours, if possible.

• Identify the jobs that use significant system resources and arrange a schedule to minimize job interference. For example, schedule jobs that are heavy disk users so that they do not run at the same time.

• Prioritize system usage.

Identify inefficient use of system resources by users or user-caused errors.

• Monitor the system for runaway processes.

• Identify inefficient user PATH variables.

• Identify heavily loaded resources using the timex command.

Detecting runaway processes

The ps(1) command provides information about active processes and light-weight processes (LWPs). (Note that many processes have only one LWP.) It gives a snapshot of what is going on, and you can use it to identify processes and LWPs that are loading the system. Examine the entries for TIME (the minutes and seconds of CPU time used by processes and LWPs) and STIME (the time when the process or LWP first started).

When you identify a runaway process or LWP (one that uses progressively more system resources over a period of time), check with the owner. A runaway process can continue to eat up system resources until everything stops. It is possible that such a process or LWP should be terminated immediately using the kill(1) command.

When you identify processes or LWPs that execute for a long time, suggest that users use the cron(1M) or at(1) commands to execute the job during off-hours.

Monitoring user PATH variables

A PATH environment variable set in each user's .profile file defines a list of directories to be searched when a command is executed. The system searches every directory in $PATH before displaying a not found error message. These searches require both processor and disk time. If there is a disk or processor slowdown, changes to PATH might help performance.

You can inform system users that inefficient PATH variables use excessive system resources, provide a default PATH, and inform users of the best way to customize their PATH variable.

Include the system directories accessible by all users in the default system PATH in the /etc/profile file.

Check PATH variables for the following:

• Since $PATH is read from left to right, the first directories specified should be those in which the most commonly used commands are located.

• Duplicate entries should not be present, so the same directory is not searched more than once for a command.

• For security reasons to reduce the chance of executing a malicious program in place of a system program, system directories should be listed first in both user $PATH variables and in the default $PATH in /etc/profile.

• If the current directory is included in the path, it should be listed last.

• In general, $PATH should have the least number of required entries.

• Searches of large directories should be avoided, but if they must be included, put them at the end of $PATH.

• A directory that is actually a symbolic link to another directory should not appear in $PATH. For example, /bin should not be in $PATH, but /usr/bin should be included.

Identifying heavily loaded system resources

The timex(1) command is used to collect information about system consumption. The results can be used to identify heavily loaded resources so that you can tune the system for better throughput. See ``Reporting application turnaround with timex'' for more information about the timex command.