Using and Administering
The tasks described in this chapter are those that you, as a user might be
interested in accomplishing and are presented in a typical step-by-step
scenario. You do not have to follow the steps shown here and may
perform certain tasks before others without any difficulty. Some tasks
must be performed prior to others in order for succeeding tasks to
work. For example, you cannot submit a job if you do not have a job
command file that you built using either the GUI or an editor.
From the Jobs window:
- SELECT
- File > Build a Job > Parallel
&TRIANGLE. The dialog box shown in Figure 33 appears:
Figure 33. LoadLeveler Build a Job Window
View figure.
Complete those fields for which you want to override what is currently
specified in your skel.cmd defaults file. A sample
skel.cmd file is found in
/usr/LoadL/full/samples. You can update this file to define
defaults for your site, and then update the *skelfile resource in
Xloadl to point to your new skel.cmd file.
If you want a personal defaults file, copy skel.cmd to one
of your directories, edit the file, and update the *skelfile
resource in .Xdefaults.
| Field
| Input
|
| Executable
| Name of the program to run. It must be an executable file.
Optional. If omitted, the command file is executed as if it were a
shell script.
|
| Arguments
| Parameters to pass to the program.
Required only if the executable requires them.
|
| Stdin
| Filename to use as standard input (stdin) by the program.
Optional. The default is /dev/null.
|
| Stdout
| Filename to use as standard output (stdout) by the program.
Optional. The default is /dev/null.
|
| Stderr
| Filename to use as standard error (stderr) by the program.
Optional. The default is /dev/null.
|
| Initialdir
| Initial directory. LoadLeveler changes to this directory before
running the job.
Optional. The default is your current working directory.
|
| Notify User
| User id of person to notify regarding status of submitted job.
Optional. The default is your userid.
|
| StartDate
| Month, day, and year in the format mm/dd/yy. The job will not
start before this date.
Optional. The default is to run the job as soon as possible.
|
| StartTime
| Hour, minute, second in the format hh:mm:ss. The job
will not start before this time.
Optional. The default is to run the job as soon as possible.
If you specify StartTime but not StartDate, the
default StartDate is the current day. If you specify
StartDate but not StartTime, the default StartTime is
00:00:00. This means that the job will start as soon as
possible on the specified date.
|
| Priority
| Number between 0 and 100, inclusive.
Optional. The default is 50.
This is the user priority. For more information on this priority,
refer to Setting and Changing the Priority of a Job.
|
| Image size
| Number in kilobytes that reflects the maximum size you expect your
program to grow to as it runs.
Optional.
|
| Class
| Class type. The job will only run on machines that support the
specified class type. Your system administrator defines the class
types.
Optional. You can press the Choices button to get a list of
available classes. Press the Details button under the class list to
verify your permissions.
|
| Hold
| Hold status of the submitted job. Permitted values are:
- user
- user hold
- system
- system hold (only valid for LoadLeveler administrators)
- usersys
- user and system hold (only valid for LoadLeveler administrators)
Optional. The default is a no-hold state. Press the set button to this field.
|
| Account Number
| Number associated with the job. For use with the llacctmrg and
llsummary commands for acquiring job accounting data.
Optional. Required only if the ACCT keyword is set to
A_VALIDATE in the configuration file.
|
| Environment
| Specifies your initial environment variables when your job starts.
Separate environment specifications with semicolons.
Optional.
|
| Shell
| The name of the shell to use for the job.
Optional. If not specified, the shell used in the owner's
password file entry is used. If none is specified, /bin/sh
is used.
|
| Group
| The LoadLeveler group name to which the job belongs.
Optional.
|
| Step Name
| The name of this job step.
Optional.
|
| Node Usage
| How the node is used. Permitted values are:
- shared
- The node can be shared with other tasks of other job steps. This is
the default.
- not shared
- The node cannot be shared.
Optional. Press the Set button to set this
field.
|
| Dependency
| A Boolean expression defining the relationship between the job
steps.
Optional.
|
| Comments
| Comments associated with the job. These comments help to
distinguish one job from another job.
Optional.
|
| Note: | The fields that appear in this table are what you see when viewing the Build
a Job window. The text in these fields does not necessarily correspond
with the keywords listed in Job Command File Keywords.
|
|
See Job Command File Keywords for information on the defaults associated with these
keywords.
- SELECT
- a Job Type if you want to change the job type you selected on the Build A
Job cascading window.
Your choices are:
- Serial
- Specifies a serial job.
- Parallel
- Specifies a non-PVM parallel job.
- PVM
- Specifies a PVM parallel job.
Note that the job type you select affects the choices that are active on
the Build A Job window.
- SELECT
- a Notification option
Your choices are:
- Always
- Notify you when the job starts, completes, and if it incurs errors.
- Complete
- Notify you when the job completes. This is the default option as
initially defined in the skel.cmd file.
- Error
- Notify you if the job cannot run because of an error.
- Never
- Do not notify you.
- Start
- Notify you when the job starts.
- SELECT
- a Checkpoint option.
Your choices are:
- No
- Do not checkpoint the job. This is the default.
- User
- Yes, checkpoint the job at intervals you determine. See checkpoint for more information.
- System
- Yes, checkpoint the job at intervals determined by LoadLeveler. See
checkpoint for more information.
- SELECT
- a Restart option
Your choices are:
- No
- Do not restart the job.
- Yes
- Yes, restart the job from an existing checkpoint file when you submit the
job.
- SELECT
- Nodes (available when the job type is parallel)
&TRIANGLE. The Nodes dialog box appears.
Complete the necessary fields to specify node information for a parallel
job. Depending upon which model you choose, different fields will be
available; any unavailable fields will be greyed out. LoadLeveler
will assign defaults for any fields that you leave blank.
| Field
| Available in:
| Input
|
| Min # of Nodes
| Tasks Per Node Model and Tasks with Uniform Blocking Model
| Minimum number of nodes required for running the parallel job. For
more information, see node.
Optional. The default is one.
|
| Max # of Nodes
| Tasks Per Node Model
| Maximum number of nodes required for running the parallel job. For
more information, see node.
Optional. The default is the minimum number of nodes.
|
| Tasks per Node
| Tasks Per Node Model
| The number of tasks of the parallel job you want to run per node.
For more information, see tasks_per_node.
Optional.
|
| Total Tasks
| Tasks with Uniform Blocking Model, and Custom Blocking Model
| The total number of tasks of the parallel job you want to run on all
available nodes. For more information, see total_tasks.
Optional for Uniform, required for Custom Blocking. The default is
one.
|
| Blocking
| Custom Blocking Model
| The number of tasks assigned (as a block) to each consecutive node until
all of a job's tasks have been assigned. For more information, see
blocking
|
| Task Geometry
| Custom Geometry Model
| The task ids of each task that you want to run on each node. You
can use the "Set Geometry" button for step-by-step directions. For more
information, see task_geometry
|
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- Network (available when the job type is parallel)
&TRIANGLE. The Network dialog box appears.
Complete those fields for which you want to specify network
information. For more information, see network.
| Field
| Input
|
| MPI/LAPI
| Choose one, both, or none of these boxes to specify the MPI (Message
Passing Interface) protocol, the (LAPI Low-level Application Programming
Interface) protocol, both protocols, or neither protocol.
Optional.
|
| Adapter/Network
| Select an adapter name or a network type from the list.
Required for each protocol you select.
|
| Adapter Usage
| Specifies that the adapter is either shared or not shared.
Optional. The default is shared.
|
| Communication Mode
| Specifies the mode in which an SP switch adapter is used, and can be
either IP (internet Protocol) or US (User Space).
Optional. The default is IP.
|
| Communication Level
| Implies the amount of memory to be allocated to each window for the
corresponding protocol, and can be Low, Average, or High.
|
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- Requirements
&TRIANGLE. The Requirements dialog box appears.
Complete those fields for which you want to specify requirements.
Defaults are used for those fields that you leave blank. LoadLeveler
dispatches your job only to one of those machines with resources that matches
the requirements you specify.
| Field
| Input
|
| Architecture*
| Machine type. The job will not run on any other machine
type.
Optional. The default is the architecture of your current
machine.
|
| Operating System*
| Operating system. The job will not run on any other operating
system.
Optional. The default is the operating system of your current
machine.
|
| Disk
| Amount of disk space in the execute directory. The job will only
run on a machine with at least this much disk space.
Optional. The default is defined in your local configuration
file.
|
| Memory
| Amount of memory. The job will only run on a machine with at least
this much memory.
Optional. The default is defined in your local configuration
file.
|
| Machine(s)
| Machine name(s). The job will only run on the specified
machines.
Optional.
|
| Feature(s)
| Features. The job will only run on machines with specified
features.
Optional.
|
| LoadLeveler Version
| Specifies the version of LoadLeveler, in dotted decimal format, on the
machine where you want the job to run. For example:
2.1.0.0 specifies that your job will run on a machine
running LoadLeveler Version 2.1.0.0 or higher.
Optional.
|
| Pool
| Specifies the number associated with the pool you want to use. All
available pools listed in the administration file appear as choices.
The default is to select nodes from any pool.
|
| Requirement
| Requirements. The job will only run if these requirements are
met.
|
| Note: |
If you enter a resource that is not available, you will NOT receive a
message. LoadLeveler holds your job in the Idle state until the
resource becomes available. Therefore, ensure the spelling of your
entry is correct. You can issue llq -s jobID to
find out if you have a job for which requirements were not met.
|
|
|
*If you do not specify an architecture or operating system, LoadLeveler
assumes that your job can run only on your machine's architecture and
operating system. If your job is not a shell script that can be run
successfully on any platform, you should specify a required architecture and
operating system.
|
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- Resources
&TRIANGLE. The Resources dialog box appears.
This dialog box allows you to set the amount of defined consumable
resources required for a job step. Resources with an "*" appended to
their names are not in the SCHEDULE_BY_RESOURCES list. For more
information, see resources.
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- Preferences
&TRIANGLE. The Preferences dialog box appears.
This dialog box is similar to the Requirements dialog box, with the
exception of the Adapter choice, which is not supported as a
Preference. Complete the fields for those parameters that you want to
specify. These parameters are not binding. For any preferences
that you specify, LoadLeveler attempts to find a machine that matches these
preferences along with your requirements. If it cannot find the
machine, LoadLeveler chooses the first machine that matches the
requirements.
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- Limits
&TRIANGLE. The Limits dialog box appears.
Complete the fields for those limits that you want to impose upon your
job. If you type copy in any field, the limits in effect on
the submit machine are used. If you leave any field blank, the default
limits in effect for your userid on the machine that runs the job are
used.
| Field
| Input
|
| CPU Limit
| Maximum amount of CPU time that the submitted job can use. Express
the amount as:
[hours:[minutes:][seconds][ .fraction]
For example, 12:56:21 is 12 hours, 56 minutes, and 21
seconds.
Optional
|
| Data Limit
| Maximum amount of the data segment that the submitted job can use.
Express the amount as:
integer[.fraction][units]
where integer and fraction represent strings of up to eight
digits.
Optional
|
| Core Limit
| Maximum size of a core file.
Optional
|
| RSS Limit
| Maximum size of the resident set size. It is the largest amount of
physical memory a user's process can allocate.
Optional
|
| File Limit
| Maximum size of a file that is created.
Optional
|
| Stack Limit
| Maximum size of the stack.
Optional
|
| Job CPU Limit
| Maximum amount of CPU a single job step can use per processor.
Optional
|
| Wall Clock Limit
| Maximum amount of elapsed time for which a job can run.
Optional
|
- SELECT
- Close to return to the Build a Job dialog box.
- SELECT
- PVM to select a PVM job.
&TRIANGLE. The PVM dialog box appears.
Complete those fields for which you want to specify requirements.
Defaults are used for those fields that you leave blank.
| Field
| Input
|
| Min # of Processors
| Minimum number of processors required for running the PVM job.
Optional. The default is one.
|
| Max # of Processors
| Maximum number of processors required for running the PVM job.
Optional. The default is one.
|
| Parallel Path
| The directory that defines where the PVM3 executables are located.
|
| PVM
| Specifies that an adapter is used for this PVM job.
|
| Adapter/Network
| Select an adapter name or a network type from the list.
Required.
|
| Adapter Usage
| Specifies that the adapter is either shared or not shared.
Optional. The default is shared.
|
- SELECT
- Close to return to the Build a Job dialog box.
There are several ways that you can edit the job command file that you just
built:
- Using the Jobs window:
- SELECT
- File > Submit a Job
&TRIANGLE. The Submit a Job dialog box appears.
- SELECT
- the job file you want to edit from the file column.
- SELECT
- Edit
&TRIANGLE. Your job command file appears in a window. You can use
any editor to edit the job command file. The default editor is
specified in your .Xdefaults file.
If you have an icon manager, an icon may appear. An icon manager is
a program that creates a graphic symbol, displayed on a screen, that you can
point to with a device such as a mouse in order to select a particular
function or application. Select this icon to view your job command
file.
- Using the Tools Edit pulldown menus on the Build a
Job window:
Using the Edit pulldown menu, you can modify the job command file.
Your choices appear in the following table:
| To
| Select
|
| Add a step to the job command file
| Add a Step
|
| Delete a step from the job command file
| Delete a Step
|
| Clear the fields in the Build a Job window
| Clear Fields
|
| Select defaults to use in the fields
| Set Field Defaults
|
| Note: | Other options include Go to Next Step, Go to Previous Step, and Go to Last
Step that allow you to edit various steps in the job command file.
|
|
Using the Tools pulldown menu, you can modify the job command
file. Your choices appear in the following table:
| To
| Select
|
| Name the job
| Set Job Name
|
| Open a window where you can enter a script file
| Append Script
|
| Fill in the fields using another file
| Restore from File
|
| View the job command file in a window
| View Entire Job
|
| Determine which step you are viewing
| What is step #
|
| Start a new job command file
| Start a new job
|
| To
| Do This
|
| Save the information you entered into a file which you can submit later
|
- SELECT
- Save
&TRIANGLE. A window appears prompting you to enter a job filename.
- ENTER
- a job filename in the text entry field.
- SELECT
- OK
&TRIANGLE. The window closes and the information you entered is saved in
the file you specified.
|
| Submit the program immediately and discard the information you entered
|
- SELECT
- Submit
- GO TO
- Step 4
|
If you already submitted your job, go to Step 4: Display, Refresh and Obtain Job Status. Otherwise, go to Step 3: Submit a Job Command File.
After building a job command file, you can submit it to one or more
machines for processing. In addition to scripts with LoadLeveler
keywords, you can also submit scripts that contain NQS options. You
cannot, however, in this release of LoadLeveler, combine NQS and LoadLeveler
options.
To submit a job, from the Jobs window:
- SELECT
- File > Submit a Job
&TRIANGLE. The Submit a Job dialog box appears.
- SELECT
- the job file that you want to submit from the file column.
You can also use the filter field and the directories column to select the
file or you can type in the file name in the text entry field.
- SELECT
- Submit
&TRIANGLE. The job is submitted for processing.
You can now submit another job or you can press Close to exit the
window.
Go to the next step.
When you submit a job, the status of the job is automatically displayed in
the Jobs window. You can update or refresh this status using the Jobs
window and selecting one of the following:
- Refresh > Refresh Jobs
- Refresh > Refresh All.
To change how often the amount of time should pass before the jobs window
is automatically refreshed, use the Jobs window.
- SELECT
- Refresh > Set Auto Refresh
&TRIANGLE. A window appears.
- TYPE IN
- a value for the number of seconds to pass before the Jobs window is
updated.
Automatic refresh can be expensive in terms of network usage and CPU
cycles. You should specify a refresh interval of 120 seconds or more
for normal use.
- SELECT
- OK
&TRIANGLE. The window closes and the value you specified takes
effect.
To receive detailed information on a job:
- SELECT
- Actions > Extended Status to receive additional
information on the job. Selecting this option is the same as typing
llq -x command.
You can also get information in the following way:
- SELECT
- Actions > Extended Details
Selecting this option is the same as typing llq -x -l
command. You can also double click on the job in the Jobs window to get
details on the job.
Note: Obtaining extended status or details on multiple jobs can be
expensive in terms of network usage and CPU cycles.
- SELECT
- Actions > Job Status
You can also use the llq -s command to determine why a submitted
job remains in the Idle or Deferred state.
For more information on these states, see llq - Query Job Status.
Go to the next step.
You can specify up to two sorting options for the Jobs
window. The options you specify determine the order in which the jobs
appear in the Jobs window.
From the Jobs window:
| Action
| Select Sort >
| Type of Sort
|
| Sort jobs by the machine from which they were submitted
| Sort by Submitting Machine >
| [Primary|Secondary]
|
| Sort by owner
| Sort by Owner >
| [Primary|Secondary]
|
| Sort by the time the jobs were submitted
| Sort by Submission Time >
| [Primary|Secondary]
|
| Sort by the state of the job
| Sort by State >
| [Primary|Secondary]
|
| Sort jobs by their user priority (last job listed runs first)
| Sort by Priority >
| [Primary|Secondary]
|
| Sort by the class of the job
| Sort by Class >
| [Primary|Secondary]
|
| Sort by the group associated with the job
| Sort by Group >
| [Primary|Secondary]
|
| Sort by the machine running the job
| Sort by Running Machine >
| [Primary|Secondary]
|
| Sort by dispatch order
| Sort by Dispatch Order >
| [Primary|Secondary]
|
| Not specify a sort
| No Sort
| [Primary|Secondary]
|
Each sorting option contains a cascading window which allows you to select
this option as either a Primary or Secondary sorting option. For
example, suppose you select Sort by Owner as the primary sorting option and
Sort by Class as the secondary sorting option. The Jobs window is
sorted by owner and, within each owner, by class.
Go to the next step.
If your job has not yet begun to run and is still in the
queue, you can change the priority of the job in relation to your other jobs
in the queue that belong to the same class. This only affects the user
priority of the job. For more information on this priority, refer to Setting and Changing the Priority of a Job. Only the owner of a job or the LoadLeveler
administrator can change the priority of a job.
From the Jobs window:
- SELECT
- a job by clicking on it with the mouse
- SELECT
- Actions > Priority
&TRIANGLE. A window appears.
- TYPE IN
- a number between 0 and 100, inclusive, to indicate a new priority.
- SELECT
- OK
&TRIANGLE. The window closes and the priority of your job changes.
Go to the next step.
Only the owner of a job or the LoadLeveler administrator can place a hold
on a job.
From the Jobs window:
- SELECT
- the job you want to hold by clicking on it with the mouse
- SELECT
- Actions > Hold
&TRIANGLE. The job is put on hold and its status changes in the Jobs
window.
Go to the next step.
Only the owner of a job or the LoadLeveler administrator can release a hold
on a job.
From the Jobs window:
- SELECT
- the job you want to release by clicking on it with the mouse
- SELECT
- Actions > Release from Hold
&TRIANGLE. The job is released from hold and its status is updated in the
Jobs window.
Go to the next step.
Only the owner of a job or the LoadLeveler administrator can cancel a
job.
From the Jobs window:
- SELECT
- the job you want to cancel by clicking on it with the mouse
- SELECT
- Actions > Cancel
&TRIANGLE. A warning dialog box appears prompting you to confirm your
cancellation request. Once you confirm your request, LoadLeveler
cancels the job and the job information disappears from the Jobs
window.
Go to the next step.
The status of the machines is automatically displayed in the Machines
window. You can update or refresh this status using the Machines window
and selecting one of the following:
- Refresh > Refresh Machines
- Refresh > Refresh All.
To specify an amount of time to pass before the Machines window is
automatically refreshed, from the Machines window:
- SELECT
- Refresh > Set Auto Refresh
&TRIANGLE. A window appears.
- TYPE IN
- a value for the number of seconds to pass before the Machines window is
updated.
Automatic refresh can be expensive in terms of network usage and CPU
cycles. You should specify a refresh interval of 120 seconds or more
for normal use.
- SELECT
- OK
&TRIANGLE. The window closes and the value you specified takes
effect.
Go to the next step.
You can specify up to two sorting options for the Machines
window. The options you specify determine the order in which machines
appear in the window.
From the Machines window:
| Action
| Select Sort >
| Sort Type
|
| Sort by machine name
| Sort by Name >
| [Primary|Secondary]
|
| Sort by schedd state
| Sort by Schedd >
| [Primary|Secondary]
|
| Sort by total number of jobs scheduled
| Sort by InQ >
| [Primary|Secondary]
|
| Sort by number of running jobs scheduled by this machine
| Sort by Act >
| [Primary|Secondary]
|
| Sort by startd state
| Sort by Startd >
| [Primary|Secondary]
|
| Sort by the number of jobs running on this machine
| Sort by Run >
| [Primary|Secondary]
|
| Sort by load average
| Sort by LdAvg >
| [Primary|Secondary]
|
| Sort by keyboard idle time
| Sort by Idle >
| [Primary|Secondary]
|
| Sort by hardware architecture
| Sort by Arch >
| [Primary|Secondary]
|
| Sort by operating system type
| Sort by OpSys >
| [Primary|Secondary]
|
| Not specify a sort
| No Sort
| [Primary|Secondary]
|
Each sorting option contains a cascading window which allows you to select
this option as either a Primary or Secondary sorting option. For
example, suppose you select Sort by Arch as the primary sorting option and
Sort by Name as the secondary sorting option. The Machines window is
sorted by by hardware architecture, and within each architecture type, by
machine name.
Go to the next step.
The LoadLeveler administrator designates one of the nodes in the
LoadLeveler cluster as the central manager. When jobs are submitted at
any node, the central manager is notified and decides where to schedule the
jobs. In addition, it keeps track of the status of machines in the
cluster and the jobs in the system by communicating with each node.
LoadLeveler uses this information to make the scheduling decisions and to
respond to queries.
To find the location of the central manager, from the Machines
window:
- SELECT
- Actions > Find Central Manager
&TRIANGLE. A message appears in the message window declaring on which
machine the central manager is located.
Go to the next step.
Public scheduling machines are those machines that participate in the
scheduling of LoadLeveler jobs on behalf of the submit-only machines.
To get a list of these machines in your cluster, use the Machines
window:
- SELECT
- Actions > Find Public Scheduler
&TRIANGLE. A message appears displaying the names of these machines.
Go to the next step.
Normally, only your jobs appear in the Jobs window.
You can, however, specify which jobs you want to appear by using the Select
pull-down menu on the Jobs window.
| To Display
| Select Select >
|
| All jobs in the queue
| All
|
| All jobs belonging to a specific user (or users)
| By User
&TRIANGLE. A window appears prompting you to enter the user IDs whose jobs
you want to view.
|
| All jobs submitted to a specific machine (or machines)
| By Machine
&TRIANGLE. A window appears prompting you to enter the machine names on
which the jobs you want to view are running.
|
| All jobs belonging to a specific group (or groups)
| By Group
&TRIANGLE. A window appears prompting you to enter the LoadLeveler group
names to which the jobs you want to view belong.
|
| All jobs having a particular ID
| By Job Id
A dialog box prompts you to enter the id of the job you want to
appear. This ID appears in the left column of the Jobs window.
Type in the ID and press OK.
|
| Note: |
When you choose By User, By Machines, or By Group, you can use a UNIX
regular expression enclosed in parentheses. For example, you can enter
(^k10) to display all machines beginning with the characters
"k10".
|
|
- SELECT
- Select > Show Selection to show the selection
parameters.
Go to the next step.
You can specify which machines will appear in the Machines
window. The default is to view all of the machines in the LoadLeveler
pool.
From the Machines window:
| To
| Select Select >
|
| View all of the machines
| All
|
| View machines by operating system
| by OpSys
&TRIANGLE. A window appears prompting you to enter the operating system of
those machines you want to view.
|
| View machines by hardware architecture
| by Arch
&TRIANGLE. A window appears prompting you to enter the hardware
architecture of those machines you want to view.
|
| View machines by state
| by State
&TRIANGLE. A cascading pulldown menu appears prompting you to select the
state of the machines that you want to view.
|
- SELECTt
- Select > Show Selection to show the selection
parameters.
Go to the next step.
Normally, all the messages that LoadLeveler generates
appear in the Messages window. If you would also like to have these
messages written to a file, use the Messages window.
- SELECT
- Actions > Start logging to a file
&TRIANGLE. A window appears prompting you to enter a filename in which to
log the messages.
- TYPE IN
- the filename in the text entry field.
- SELECT
- OK
&TRIANGLE. The window closes.
[ Top of Page | Previous Page | Next Page | Table of Contents | Index ]
View figure.