Region Batch Processing Interface

Overview

Several Priism applications which use a command-line interface also have a graphical interface to specify the command-line arguments and some other aspects (priority, batch queue) of how the application will be run. The elements that these graphical interfaces have in common and the corresponding command-line arguments are described in this document; application specific elements are described elsewhere.

At the top of the interface are entries for the input and output files for the application. For each file there is a button which you can use to open a file selection dialog and a text field which shows the current file selection. You can change the selected file by changing the name shown in the field.

Below the file selection items, some applications display a series of controls to select a subset of the input to process. These controls are further described in the X Range, Y Range, Z Range, T Range, and Wave topics. Other applications simply display the size of the input data.

The remainder of the dialog, with the exception of the buttons at the bottom, contains items specific to the application. The functions of the buttons at the bottom are:

Exit: Closes the dialog.
Save: Saves the current dialog settings to a file in your home directory. You can reload these settings at a later time using the Restore button.
Restore: Loads the dialog entries that had been previously saved using the Save button.
Options: Opens a dialog which allows you to control how the application will be run including whether or not a batch queue should be used, what priority the job should have, names for the command and script files, and whether or not to chain multiple jobs together. For more information see the Priority, Chain, Queue, Logfile, and Comfile topics.
Do It: If the option to chain jobs is turned off (this is set from the dialog opened by the Options button), pressing this button causes the application to be run with the current settings. If the application is run on the current machine, the label on the button will be changed to Cancel while the job runs and pressing the button will terminate the job.
The first time the Do It button is pressed after the option to chain jobs is turned on, nothing will be executed though the commands that would have been executed will be written to the command file. The next time the "Do It" button is pressed, the commands to run the current job are written to a command file, a line is added to the previous command file to run the one with the commands for the current job, and the previous command file is executed. The label on the Do It button then changes to Cancel and the button can be used to cancel the combined jobs.

Topics

X Range

Next to the X Range label is a field which shows the full x size of the input data in the form

            <minimum x index>:<maximum x index>

Next to that field is another in which you can specify the x range to process. The first value in the field is the first x index to process and the second value is the last x index to process.

On the command-line, the x range to process is specified with an option of the form (optional parts are listed in brackets)

            -x=xstart[:xend]

If the option is not supplied, the entire x range of the input is processed. When the last x index to process, xend, is omitted, it is assumed to be the maximum x index in the input file.

Topics

Y Range

Next to the Y Range label is a field which shows the full y size of the input data in the form

            <minimum y index>:<maximum y index>

Next to that field is another in which you can specify the y range to process. The first value in the field is the first y index to process and the second value is the last y index to process.

On the command-line, the y range to process is specified with an option of the form (optional parts are listed in brackets)

            -y=ystart[:yend]

If the option is not supplied, the entire y range of the input is processed. When the last y index to process, yend, is omitted, it is assumed to be the maximum y index in the input file.

Topics

Z Range

Next to the Z Range label is a field which shows the full z size of the input data in the form

            <minimum z index>:<maximum z index>

Next to that field is another in which you can specify the z range to process. The first value in the field is the first z index to process and the second value is the last z index to process.

On the command-line, the z range to process is specified with an option of the form (optional parts are listed in brackets)

            -z=zstart[:zend]

If the option is not supplied, the entire z range of the input is processed. When the last z index to process, zend, is omitted, it is assumed to be the maximum z index in the input file.

Topics

T Range

Next to the T Range label is a field which shows the full time extent of the input data in the form

            <minimum time index>:<maximum time index>:1

Next to that field is another in which you can specify the time range to process. The first value in the field is the first time index to process, the second value is the last time index that could be processed, and the last value is the index step between times that are processed.

On the command-line, the time range to process is specified with an option of the form (optional parts are listed in brackets)

            -t=tstart[:tend[:tstep]]

If the option is not supplied, the entire time range of the input is processed. When the time index step, tstep is omitted, it is assumed to be one. When the last time to process, tend is omitted, it is assumed to be the maximum time index in the input file.

Topics

Wave

The wavelengths available in the input are shown next to the Wave label. Those wavelengths that will be included in the processing have the toggle button next to them in the on state.

On the command line, the wavelengths to be processed are specified with an option of the form (optional parts are listed in brackets)

            -w=wave1[:wave2[:wave3[:wave4[:wave5]]]]

When the option is not specified, all input wavelengths are processed. A wavelength can either be listed by its wavelength or by its index in the input file (a value between zero and the number of input wavelengths minus one).

Topics

Size Display

Some applications do not display the range and wavelength selection controls and the underlying command-line application does not accept the corresponding command-line arguments. Instead, the application displays the size of the current input file: either as the number of points along each of the five dimensions (x, y, z, wavelength, and time) or as the number of points along the x and y dimensions and the the number of sections.

Topics

Priority

When the Lower priority toggle is on, commands will be executed at a lower priority for access to the computer's resources. It is equivalent to using the Unix nice command with the default priority increment to decrease a command's priority.

Topics

Chain

When the Chain option toggle is off and the "Do It" button is pressed, the application is immediately sent to a batch queue or executed locally. When the toggle is on, the needed commands are written to the command file when the Do It button is pressed but are not submitted or run until the next time the Do It button is pressed. To use this effectively to chain two jobs together, do the following

Turn the chain option toggle on.
Set the parameters for the first job and press Do It.
Change the name of the command file; using the same name as was used for the first job will cause the first job to run in an infinite loop.
Set the parameters for the second job and press Do It.
Turn the chain option toggle off.

Because submitting the second job turns the Do It button into a button to cancel the currently running job, it is not generally possible to chain more than two jobs together.

Topics

Skip Execution

When the "Skip execution" button is off, pressing the "Do It" button generates a script of the commands to perform and, depending on the queue selected, either immediately starts executing the commands or submits the job to a queue. Using the chain option can modify that sequence of events slightly to run two tasks in sequence.

When the "Skip execution" button on, pressing the "Do it" button only generates a script of the commands to perform. In this case, the selection of the queue to use generally does not matter. One exception to that rule is parallel tasks since the default parallel execution parameters are what has been configured for the currently selected queue.

Topics

Queue

There is the option to run the application in a batch queue. Use the pulldown menu next to "Submit to queue" to select the batch queue to use or use the run locally option to run jobs on the current machine without using a queue.

There are two ways to configure the queues listed in the dialog: you can set up a configuration file or you can set some environment variables. The former is more flexible and allows you to configure parallel execution settings that can not be configured with the environment variables. A configuration file, if present, takes precedence over the environment variable settings. Priism looks for configuration files in four places and uses the first it finds. The locations, in the order Priism searches them, are:

home/.iveprefs/host-queues.def
home/.iveprefs/queues.def
priism/CONFIG/host-queues.def
priism/CONFIG/queues.def

where home is the user's home directory, priism is the Priism installation directory, and host is the name, as reported by the command "uname -n", of the machine where Priism is running.

Queue Configuration Files

EXAMPLE/queues.def in the Priism distribution is an example of a queue configuration file. You may want to look at it first before plunging into the details below.

The configuration file consists of a series of entries where each entry is separated from the previous one by an end-of-line. Entries can be preceded by white space (spaces or horizontal tabs) and can be followed by whitespace or a comment (a comment is introduced with a # character and continues until the end of the line). An entry is either a directive which controls how the file is parsed or the definition of a property.

There are five directives available. They are:

HOST: Introduces a set of definitions for the machine where Priism is running. All properties up to the next HOST directive, QUEUE directive, or the end of the file are grouped together.
QUEUE: Introduces the definitions for a new destination queue. All properties up to the next HOST directive, QUEUE directive, or the end of the file are applied to that queue.
IFHOST: If the host name of the current machine is not included in the list of hosts following the IFHOST directive, all properties up to the matching ENDIF directive are read but ignored.
IFNOTHOST: If the host name of the current machine is included in the list of hosts following the IFHOST directive, all properties up to the matching ENDIF directive are read but ignored.
ENDIF: Terminates a previous IFHOST or IFNOTHOST directive.

As a consequence of the way the QUEUE and HOST directives are defined, properties that are to be used for the entire system must appear before the first QUEUE or HOST directives.

The IFHOST and IFNOTHOST directives have an optional list of host names. The list of host names must be separated from the IFHOST or IFNOTHOST directive by white space. The list is introduced by a { character and terminated by a } character (} characters within host names do not terminate the list). Each element in the list is enclosed in double quotes and the elements may be separated from each other by any combination of white space, comments (which start with a # and continue till the end of the line; the list will continue on the next line), or an end-of-line (the list will continue on the next line). If you want to include a literal double quote in a host name, use a backslash followed by a double quote. If you want to include a literal backslash at the end of a host name use two backslashes followed by the double quote. If an IFHOST directive does not have a list of host names or the list is empty, the properties up to the terminating ENDIF are always excluded. If an IFNOTHOST directive does not have a list of host names or the list is empty, the properties up to the terminating ENDIF are always included.

Properties consist of a name followed by an optional value. Names consist of a sequence of characters (not including white space, # characters, or end-of-lines) followed by white space, a # character, or an end-of-line. A name must not be the same as one the of directives (if it is, it will be processed as a directive). There are four types of values:

boolean: The value is either true or false.
integer: The value is a signed decimal integer (in other words an optional plus or minus sign followed by a sequence of digits).
character string: The value is introduced with a double quote (which is not included in the value itself) and continues until the next double quote (which does not have to be on the same line). If you want to include a double quote in the value, use a backslash followed by a double quote. If you want to include a backslash at the end of the value use a pair of backslashes followed by the double quote.
a list of character strings: The list is introduced by a { character and terminates with a } character (} characters within strings in the list do not terminate the list). Each element in the list is enclosed in double quotes and the elements may be separated from each other by any combination of white space, comments, or end-of-lines. If you want to include a double quote in the value, use a backslash followed by a double quote. If you want to include a backslash at the end of the value use a pair of backslashes followed by the double quote.

Properties without a value are treated as if an empty character string had been supplied as the value.

Properties are distinguished by three attributes: the name, the type of value, and the enclosing context. The enclosing context is introduced by a HOST or QUEUE directive; any properties before the first or HOST or QUEUE directive belong to the same context and are used as the properties of the overall system. If the three attributes are the same for two properties then one of the following applies:

If the property's value is a boolean, integer, or character string, the property that appears later in the file takes precedence.
If the property's value is a list of character strings, the lists are merged and the elements in the combined list will appear in the order they do in the file.

While any combination of the three attributes is possible only a handful are recognized by the software and affect the graphical user interfaces or submission of jobs. For the overall system properties, four have special meanings (the name of the property is show first followed by the expected type in parentheses):

DefaultSerialQueue (character string): Holds the name of the default destination queue for serial (i.e. single processor) tasks. If the character string is empty or there is no queue with the given name, the default destination for serial jobs is the local machine.
DefaultThreadedQueue (character string): Holds the name of the default destintation queue for parallel tasks that can only run on something that is (or acts like) a single host. If the character string is empty or there is no queue with the given name, the default destination for parallel tasks that use a single host is the local machine.
DefaultClusterQueue (character string): Holds the name of the default destination queue for parallel tasks that may run on multiple hosts. If the character string is empty or there is no queue with the given name, the default destination for parallel tasks that may run on multiple hosts is the local machine.
DefaultGPUQueue (character string): Holds the name of the default destination queue for GPU-accelerated tasks. If the character string is empty or there is no queue with the given name, the default destination for GPU-accelerated jobs is the local machine.

For the host properties, the following have special meanings:

DefaultBackgroundLoginOption (character string): Holds the option to pass to the login command when logging in to a node to run a task in the background. This property defaults to "-n" if it is not set.
DefaultLoginCommand (character string): Holds the command to use to login to a node. This property defaults to "ssh -x" if not set.
DefaultMachinesFile (character string): This is only relevant for queues or hosts where ProcessControl is "mpich-machines". Holds the name of the MPICH-style machines file with the list of processor names to use.
DefaultMPIRunCommand (character string): Holds the command necessary for starting an MPI job. For systems or queues that set ProcessControl to "openmpi", the command must be OpenMPI's (version 1.3 or later) mpirun or mpiexec.
DefaultNodeCheckCommand (character string): Holds the command to be executed before attempting to log in and run a task when ProcessControl is "mpich-machines" or "openmpi". The intent of this is to work around problems with intermittent failures. For instance, on some of the clusters we had there were occasional NFS mount failures as the parallel script runs; using a non-empty DefaultNodeCheckCommand can be used to insert a delay or a normal login to the compute node to prevent those failures.
DefaultProcessorCount (integer): Holds the default number of processors to use for parallel jobs.
Description (character string): Holds a longer description of the system for use in the user interface.
GrecsrvBusyHosts (character string): Holds a comma-separated list of device IDs to ignore on each host.
GrecsrvExecutable (character string): Holds the default path to GPU reconstruction server software.
GrecsrvHosts (character string list): Lists the host name or address for hosts to use with the GPU reconstruction server software. This would usually be used when the GrecsrvUseProcessControl property is false.
GrecsrvMemoryPercentatge (integer): Holds the percentage of each device's onboard memory that the GPU reconstruction server will attempt to use.
GrecsrvPort (integer): Holds the port number to use for GPU reconstruction server hosts that do not otherwise have a port number specified.
GrecsrvRunning (boolean): If true, jobs will assume that the GPU reconstruction server software is already running. Otherwise, jobs will start and stop the GPU reconstruction server software. If not set, this property defaults to be false.
GrecsrvUseProcessControl (boolean): If true, jobs will use the mechanism described by the ProcessControl property to determine the hosts that will run the GPU reconstruction server software and the mechanism for starting the software. If false, jobs will use the hosts listed by the GrecsrvHosts property and will use the login command set by the DefaultLoginCommand property to start the server on those hosts. If not set, this property defaults to be true.
ProcessControl: Defines how parallel jobs that start their own child processes determine where the processes should run and implement the procedure of starting the child process. Three values are currently understood: "simple", "mpich-machines", and "openmpi". "simple" means that all the processes should be started on the same host as the parent process. "mpich-machines" means that the job will be passed a text file, in the same format as used by MPICH for its machines files, listing the hosts to use. The job will also be passed the command to use to log in to other hosts. The DefaultBackgroundLoginOption, DefaultLoginCommand, DefaultMachinesFiles, and DefaultNodeCheckCommand properties are useful for systems or queues that use "mpich-machines" for ProcessControl. "openmpi" means that the job will be passed how to invoke OpenMPI's (version 1.3 or later) mpirun or mpiexec and should use that when starting processes. The DefaultMPIRunCommand is useful for systems or queues that use "openmpi" for ProcessControl.
StatusCommand (character string): Holds the command line to query the system's status for display in the graphical user interface.

The queue properties with special meaning include all the host properties and the following properties:

AcceptsCluster (boolean): Indicates whether or not the queue accepts parallel jobs that may run on multiple hosts. If this property is not set for a queue, the default value is true.
AcceptsGPU (boolean): Indicates whether or not the queue accepts GPU-accelerated jobs. If this property is not set for a queue, the default value is false.
AcceptsSerial (boolean): Indicates whether or not the queue accepts serial (single-processor) jobs. If this property is not set for a queue, the default value is true.
AcceptsThreaded (boolean): Indicates whether or not the queue accepts parallel jobs that run on a single host. If this property is not set for a queue, the default value is true.
Name (character string): Holds the name of the queue as will appear in the menus for selecting queues.
SetProcessorCountForApp (boolean): If true the requested number of processors is passed to the application. If false, the requested number of processors is only passed to the command given by SubmissionCommand. When not set, this property defaults to be true.
SubmissionCommand (character string): Holds the command necessary for submitting a job to the queue. The command followed by white space followed by the name of the script or executable to run must work to start a job. Occurrences of %p in the command will be replaced by the number of processors to use. Occurrences of %% in the command will be replaced by a single % character.

Queue Environment Variables

If you use environment variables rather than a configuration file to specify the queues, the environment variables QUE1, QUE2, ..., QUE10 set the names of the batch queues. The command to submit a task to a queue can be set in one of two ways. To set a command specific to a single queue, use the QUECOM1, ..., or QUECOM10 variables. The command combined with the name of the command file (for instance,

            $QUECOM1 script.com

) is what is used when submitting a task. Use the QCOMMAND environment variable to set a command for all queues that do not have a specific command set. When QCOMMAND is used,

            $QCOMMAND <name of queue> <name of script file>

is what is used when submitting a task.

By default, jobs are run on the local machine. If you want one of the queues to be used by default for single processor jobs, set the QDEFAULT environment variable to the number of the queue which is to be the default destination. A value of QDEFAULT which is not between one and ten will set the default destination to the local machine.

Topics

Logfile

Anything that the application writes to standard output and standard error will be stored in the file listed in the LogFile field. If the LogFile field is blank, standard output and standard error are not redirected by the command file.

Topics

Comfile

The commands used to run the application are written to the file listed in the ComFile field. If you intend to run multiple jobs at once (for instance if you are submitting multiple jobs to batch queues), you should change the name of the command file for each one so that you don't overwrite the commands for a previous job that may not have completed. It would also be a good idea to change the name of the log file for each job as well.

Topics

Overview

Topics

Related Priism Topics

Topics

Topics

Topics

Topics

Topics

Topics

Topics

Topics

Topics

Queue Configuration Files

Queue Environment Variables

Topics

Topics

Topics