IBM LoadLeveler for AIX 5L: Using and Administering

LoadLeveler basics

LoadLeveler has three types of interfaces that enable users to create and submit jobs and allow system administrators to configure the system and control running jobs. These interfaces include:

1. Command line interface
   • The command line interface gives you access to basic job and administrative functions.
   • For more information see: LoadLeveler command line interface
2. A Graphical User Interface (GUI)
   • LoadLeveler's GUI provides system access similar to the command line interface. Experienced users and administrators may find the command line interface more efficient than the GUI for job and administrative functions.
   • For more information see: Using the Graphical User Interface
3. Application Programming Interface (API)
   • The Application Programming Interface allows application programs written by users and administrators to interact with the LoadLeveler environment.
   • For more information see: LoadLeveler API interface

All three types of interfaces permit different levels of access to users and administrators. User access is typically restricted to submitting and managing individual jobs, while administrative access allows setting up system configurations, job scheduling, and accounting.

• For background information on user and administrator tasks see Network job management and job scheduling systems
• For detailed information on user tasks see:
  • Submitting and managing jobs
  • Special considerations for parallel jobs
• For detailed information on administrator tasks see:
  • Administering and configuring LoadLeveler
  • Administration tasks for parallel jobs
  • Gathering job accounting data
  • Routing jobs to NQS machines

Using either the command line or the Graphical User Interface, users create job command files that instruct the system on how to process information. Each job command file consists of keywords followed by the user defined association for that keyword. For example, the keyword "executable" tells LoadLeveler that you are about to define the name of a program you want to run. Therefore, "executable = longjob" tells LoadLeveler to run the program called "longjob."

After creating the job command file, you invoke LoadLeveler commands to monitor and control the job as it moves through the system. LoadLeveler monitors each job as it moves through the system using process control daemons. However, the administrator maintains ultimate control over all LoadLeveler jobs by defining job classes that control how and when LoadLeveler will run a job.

For more information on job command files see:

• Job definition
• Submitting and managing jobs

For more information on keywords used in job command files see:

• Building a job command file
• Keyword considerations for parallel jobs
• Job command file keywords
• Administration and Configuration file keywords

For more information on commands used in job command files see:

• Submitting a job command file
• Managing jobs
• Commands

For more information on daemons see:

• LoadLeveler daemons
• LoadLeveler daemons and job states

For more information on job classes see:

• How LoadLeveler schedules jobs
• Administration file structure and syntax
• Setting up a class for parallel jobs

In addition to setting up job classes, the administrator can also control how jobs move through the system by specifying the type of scheduler. LoadLeveler has several different scheduler options that start jobs using specific algorithms to balance job priority with available machine resources. For more information on scheduler options see:

• How LoadLeveler schedules jobs
• Scheduler considerations
• Scheduling considerations for parallel jobs
• Choosing a scheduler

When LoadLeveler administrators are configuring clusters and when users are planning jobs, they need to be aware of the machine resources available in the cluster. These resources include items like the number of CPUs and the amount of memory available for each job. Because resource availability will vary over time, LoadLeveler defines them as consumable resources. For more information on consumable resources see:

• Consumable resources
• Managing jobs that consume resources
• Gathering job accounting data