     Loose Integration Between Grid Engine and Sun HPC ClusterTools(TM)
     ------------------------------------------------------------------

Content
-------

1. Introduction
2. Integration Procedure
3. Technical Details for Loose Integration
   3.1. Exclusive Complex and Queues for Sun HPC ClusterTools
   3.2. Retrieval of the CRE Job Id
   3.3. Start scripts
   3.4. Stop scripts
   3.5. Suspend/Resume/Terminate methods
   3.6. MPRUN script
   3.7. PRISM script
   3.8. Grid Engine batch script
   3.9. ClusterTools User Access
4. How to Add/Modify Grid Engine Objects
   4.1. Exclusive complex object
   4.2 Queue objects
   4.3. PE object
   4.4. User Access List (ACL) object
5. Limitations in the Current Loose Integration
6. Test for the Loose Integration
   6.1. Batch script for HPC ClusterTools
   6.2. Interactive command for HPC ClusterTools
   6.3. Prism debugging job
Appendix A: Step by Step Loose Integration
      Step 1: Create appropriate scripts for the loose integration
   A.1. MPRUN: a wrapper for the mprun command of HPC ClusterTools
   A.2. PRISM: a wrapper for the prism command of HPC ClusterTools
   A.3. startsunmpi.sh: a script to prepare a rank map file for HPC ClusterTools
   A.4. stopsunmpi.sh: a script to be executed at the end of MPI jobs
   A.5. suspend_sunmpi.sh: a script to suspend MPI jobs
   A.6. resume_sunmpi.sh: a script to resume a suspended MPI job
   A.7. terminate_sunmpi.sh: a script to terminate MPI jobs
      Step 2: Create an exclusive complex for HPC ClusterTools
      Step 3: Create exclusive queues for Sun HPC ClusterTools
      Step 4: Create a parallel environment for HPC ClusterTools
Copyright


1. Introduction
---------------

This package is to create a loose integration between Sun HPC ClusterTools
(Versions 3.1 & 4.0) and Grid Engine (including Enterprise Edition) 5.3. 
The package is provided to users as an unsupported tool and should be seen
as an advanced starting point for a site specific customization. The package
intends to help Sun HPC ClusterTools and Grid Engine users so that they can
set up the loose integration with a little effort. However, it should be
noted that the package is provided on "as is" basis and may not work with
all of Grid Engine functions. Users should be aware of the limitation of
this loose integration, which will be noted later in this note. Furthermore,
the package assumes that users are familiar with Sun HPC ClusterTools and
Grid Engine. Users should refer to the corresponding documentation for more
information about each product.

The package contains the following files:

   README (this file)
   HPC_GE_Integration.sh
   startsunmpi.sh
   stopsunmpi.sh
   suspend_sunmpi.sh
   resume_sunmpi.sh
   terminate_sunmpi.sh
   MPRUN
   PRISM
   batch.sunmpi.template
   sunmpi.pe.template

The integration script (HPC_GE_Integration.sh) assumes that both Sun HPC
ClusterTools (Versions 3.1 & 4.0) and Grid Engine (including Enterprise
Edition) 5.3 are installed in advance before executing the script. The
script will create a Parallel Environment (PE) with the "fill-up" resource
allocation scheme. The PE is required for Grid Engine to submit a parallel
job such as an MPI job. The script also does other modifications relevant to
the PE environment setup in order to use Sun HPC ClusterTools software with
Grid Engine. The start and stop scripts (startsunmpi.sh and stopsunmpi.sh)
are used to prepare a proper start-up environment for Sun HPC ClusterTools
and to finish a Sun MPI job properly. The suspend, resume and terminate
scripts (suspend_sunmpi.sh, resume_sunmpi.sh and terminate_sunmpi.sh) are
used to suspend/resume/terminate an MPI job. The MPRUN script is an mprun
wrapper to save the process Id of an mprun process into a temporary file.
This pid is used by the suspend and terminate scripts to get Cluster Runtime
Environment (CRE) job Id of an MPI job. A similar wrapper called PRISM is
also provided to start the prism debugger. The PRISM wrapper is useful for a
debugging job via Grid Engine environment. The batch.sunmpi.template is a
template for a batch job script. The sunmpi.pe.template is a template for
the parallel environment object.


2. Integration Procedure
------------------------

The integration between Grid Engine and Sun HPC ClusterTools can be done by
either executing the integration script or manually configuring appropriate
configuration files using qconf command or qmon graphical user interface.

The integration script, HPC_GE_Integration.sh, will create a loose
integration between Grid Engine and Sun HPC ClusterTools. This integration
script uses Grid Engine and HPC ClusterTools commands, qconf and mpinfo. So
appropriate environment setting for Grid Engine and HPC ClusterTools must be
done in order to access the commands before executing the integration
script.

In addition, administrator for Grid Engine should execute the integration
script to avoid any permission problem.  The script will ask a few questions
to find out the installation information. Administrator should answer all
the questions correctly according to a site-specific installation
information. The following information will be asked by the integration
script.

	Grid Engine installation root directory
	Grid Engine cluster (cell) name
	Parallel Environment (PE) name
	Exclusive complex and its shortcut name for Sun HPC ClusterTools
	List of execution hosts running Sun HPC ClusterTools
	List of exclusive queues for execution hosts running Sun HPC 
        ClusterTools
	Sun HPC ClusterTools user group name and its members	
	Parallel Environment (PE) slots (available number of CPUs)

Since the integration script uses qconf command interface, there is no need
to restart Grid Engine daemons after completing the loose integration.

If a manual setup is preferred instead of running the integration script,
one can use the appendix as a guildeline for the step-by-step procedure for
the loose integration between Grid Engine and Sun HPC ClusterTools.

If there are more than one queue defined for an execution host as a result
of adding the exclusive CRE queue, then, those queues for an execution host
should be synchronized by defining slots as consumable to the appropriate
execution host objects. The loose integration script provides an automated
procedure to customize execution hosts for queue synchronization. Also this
can be done by using "qconf -me <hostname>" command. Then, replace NONE on
the line defining "complex_values" with "slots=<available_CPUs>". Without
this synchronization, both queues will accept any batch or interactive job
submitted to the queues until it reached its own limit. Therefore, this will
eventually overload the particular execution host.


3. Technical Details for Loose Integration
------------------------------------------

3.1. Exclusive Complex and Queues for Sun HPC ClusterTools
----------------------------------------------------------

The integration script will set up an exclusive environment for MPI jobs
using Sun HPC ClusterTools while keeping the existing Grid Engine
installation unaffected. In order to create such an exclusive environment
for MPI jobs, the integration script creates exclusive queues defined as
<hostname>.cre for each execution host running HPC ClusterTools. By doing
so, it will not change any existing queues for the current Grid Engine
installation. However, since the loose integration is performed while the
Grid Engine is running, the CRE-only queues may be scheduled by the Grid
Engine scheduler for non-MPI jobs as soon as they are created. So a new
complex, which is a named container for a set of attributes in Grid Engine,
is added to the Grid Engine configuration and attached to the CRE queues in
order to prevent non-MPI jobs slipping undesirably into the CRE queues. In
this case, the new complex called cre_exclusive by default, is defined as
shown below:

#name  shortcut  type  value  relop  requestable  consumable  default
#--------------------------------------------------------------------
cre    cre       BOOL  true   ==     FORCED       NO          false

The cre_exclusive complex is attached to each CRE queues by referring to it
in the 'complex_list' and by setting the value in the 'complex_values' of
the appropriate object. For the exclusive CRE queues, the following lines
are defined:

  complex_list         cre
  complex_values       cre=true

in each CRE queue. With this configuration, Grid Engine jobs will not get
dispatched to the CRE queues if they do not request this 'cre' attribute
explicitly, e.g.

	 qsub -l cre -pe hpc 4 script.sh 

Like other submission options, the '-l cre' can be put in the Grid Engine
job script to ensure the job is not submitted without this request.


3.2. Retrieval of the CRE Job Id
--------------------------------

The most important issue in this loose integration is how to get the CRE job
Id of an MPI job. Both suspend and terminate methods defined by the loose
integration use the mpkill command of Sun HPC ClusterTools to propagate
signals to MPI jobs and the mpkill command requires the CRE job Id as an
input. The current implementation to retrieve the CRE job Id involves a few
steps. The first step is to get the process Id of an mprun process which is
the command starting an MPI job. This is done by an mprun wrapper called
MPRUN, which is distributed with the loose integration package. The mprun
process Id is saved to a temporary file, mprun_pid, located at $TMPDIR.
Then, using this process Id, the core file of the mprun process is saved to
the same $TMPDIR. Finally, the adb utility is used to search for the CRE job
Id of the corresponding MPI job. The extracting process of the CRE job Id is
done by both suspend and terminate scripts.


3.3. Start scripts
------------------

The start script for Sun MPI jobs is named as "startsunmpi.sh". It needs two 
argument input, $pe_hostfile and "/opt/SUNWhpc/bin". The first argument, 
$pe_hostfile, is a file automatically generated by Grid Engine. This file 
contains a list of host names and number of processes to be used for a
particular MPI job. The location of $pe_hostfile is dynamically determined
by Grid Engine based on its job execution. The second argument is the path
for Sun HPC ClusterTools binaries. The start script will check the path to
find out if Sun HPC ClusterTools is available. The start script will parse
$pe_hostfile and generate a rank map file called machines. This rank map file 
will be used by "mprun" with "-Mf" option or "-m" option depending on 
ClusterTools' version and be placed at $TMPDIR directory. During the run 
time, Grid Engine creates $TMPDIR for each batch job on the master execution 
host which is an execution host launching the batch script. The absolute path 
for $TMPDIR is determined by Grid Engine based on the batch job Id. After the 
loose integration, the start script will be located at $SG_ROOT/mpi/sunhpc/
loose-integration directory, where $SGE_ROOT is an absolute path for the root 
directory for Grid Engine installation.


3.4. Stop scripts
-----------------

The stop script for Sun MPI jobs is named as "stopsunmpi.sh". It will read a
file called $TMPDIR/ptree.list and kill all the processes listed in that
file. The $TMPDIR/ptree.list file is generated by a script defined by the
"terminate_method" of each CRE queue assigned to a PE. The stop script will
be located where the start script is.


3.5. Suspend/Resume/Terminate methods
-------------------------------------

The loose integration script also defines suspend, resume and terminate
methods in the exclusive CRE queues listed in the PE configured during the
integration process. Both suspend and terminate scripts (suspend_sunmpi.sh
and terminate_sunmpi.sh) are designed to pick up the CRE Job Id of an MPI
job from the mprun core image by using adb utility. It should be noted that
the CRE Job Id is different than the batch Job Id. After picking up the CRE
Job Id, both suspend and terminate scripts uses "mpkill" command to either
suspend or kill a Sun MPI job. Furthermore, the terminate script will
generate a list of all the processes related to a given MPI job and save it
to a file called $TMPDIR/ptree.list. This file will be used later to clean
up any zombie processes once the MPI job terminated. Both scripts will be
located at $SGE_ROOT/mpi/sunhpc/loose-integration directory. A resume method
(resume_sunmpi.sh) is provided to be counter part of the suspend method. The
resume method uses the CRE Job Id saved at the $TMPDIR/cre_job_id by the
suspend method.

When a batch job script launches more than one MPI jobs or contains other 
tasks in it, it is necessary to propagate the suspend/resume/terminate 
signals to the process group. This can be done by executing the corresponding 
kill command with specifying negative pid of the batch job script, which is 
passed to the suspend/resume/terminate scripts via $job_pid, right after 
issuing the mpkill command within the suspend/resume/terminate scripts. The 
mechanism to pass $job_pid to the suspend/resume/terminate scripts is 
described in the 4.2 Queue objects section.


3.6. MPRUN script
-----------------

The MPRUN script is a wrapper for the mprun command. It saves the process Id
of an mprun process into a file, mprun_pid, at a temporary directory,
$TMPDIR. The mprun process Id will be used to obtain a CRE job Id by suspend
and terminate scripts for suspension and termination of an MPI job. If a
user does not use the wrapper to start an MPI application, the current loose
integration does not get a CRE job Id. Hence, Grid Engine command such as
qmod or qdel will not be able to control any MPI jobs. Also the MPRUN
wrapper sets the CRE environment variable, MPRUN_FLAGS using "-Mf" option or
"-m" option depending on ClusterTools' version with the rank map file,
$TMPDIR/machines so that users do not need to define the rank map file in 
their job submissions.

3.7. PRISM script
-----------------

The PRISM script is a wrapper for the prism command. It sets the CRE
environment variable, MPRUN_FLAGS similar to the MPRUN wrapper so that users
do not need to define the rank map file in their job submissions. This also
make sure that the prism dubugger uses the resources allocated by the Grid
Engine.


3.8. Grid Engine batch script
-----------------------------

The loose integration package distributes an example Grid Engine batch
script called "batch.sunmpi.template". The file will be located at
$SGE_ROOT/mpi/sunhpc/loose-integration directory.  It should be noted that
an mprun wrapper, MPRUN, is used instead of "mprun" command. This is
necessary to set up a loose integration between Grid Engine and HPC
ClusterTools so that an MPI job can be controlled by Grid Engine commands
such as qmod and qdel.  Users should customize it according to their Grid
Engine installation before using it.

For some commercial MPI applications, they use the mprun command within
their start-up scripts. In this case, users may not be able to modify their
scripts to replace the mprun command with the MPRUN wrapper. One quick
workaround is to use the environment variable, PATH, to overload the mprun
command as shown below:

	cd $SGE_ROOT/mpi/sunhpc/loose-integration
	ln -s MPRUN mprun
	Place <SGE_ROOT>/mpi/sunhpc/loose-integration path at the head of $PATH

In this way, when the script  executes the mprun command, it will pick up
the MPRUN wrapper from $SGE_ROOT/mpi/sunhpc/loose-integration instead of the
actual mprun command.


3.9. ClusterTools User Access
-----------------------------

In order to use resources controlled by Grid Engine, administrator can set a
proper user access list (ACL) with its users. The integration script will
check the ACL lists and will ask whether to create an ACL or to skip this
process.


4. How to Add/Modify Grid Engine Objects
----------------------------------------

There are a number of Grid Engine objects such as complex, queue, PE, and so
on, to be added/modified for the loose integration between Grid Engine and
HPC ClusterTools. The following shows how to add or modify Grid Engine
objects for the loose integration.


4.1. Exclusive complex object
-----------------------------

The "qconf -scl" will list all complex definition objects in the Grid Engine
system. The exclusive complex for HPC ClusterTools is named as
"cre_exclusive" by default. Administrator may change the default name during
the integration setup. The contents of the exclusive complex can be viewed
using the "qconf -sc cre_exclusive" command. The command to create a complex
is "qconf -Ac <complex_name> <complex_def_file>". Also one can use "qconf
-ac <complex_name>" to add a complex or "qconf -mc <complex_name>" to modify
an existing complex interactively.


4.2 Queue objects
-----------------

The "qconf -sql" will list all queue objects in the Grid Engine system. The
exclusive CRE queues associated with execution hosts running Sun HPC
ClusterTools are named as <hostname>.cre by default. The command, "qconf -sq
<queue_name>", will show the contents of the specified queue. The
integration script creates exclusive queues for Sun HPC ClusterTools using
the command, "qconf -Aq <queue_def_file>". The following lines are the
modified lines relevant to loose integration with Sun HPC ClusterTools.

	qname               <hostname>.cre
	qtype                BATCH INTERACTIVE PARALLEL 
	suspend_method       $SGE_ROOT/mpi/sunhpc/loose-integration/suspend_sunmpi.sh $job_pid
        resume_method        $SGE_ROOT/mpi/sunhpc/loose-integration/resume_sunmpi.sh $job_pid
	terminate_method     $SGE_ROOT/mpi/sunhpc/loose-integration/terminate_sunmpi.sh $job_pid
	complex_list         cre_exclusive
	complex_values       cre=true

where $SGE_ROOT represents an absolute path for the Grid Engine
installation.

Users can also use the "qconf -mqattr" command to modify each attribute of 
the existing queues individually. For example, the following shows how to 
customize the suspend_method attribute of a queue object.

% qconf -mqattr suspend_method \
  "$SGE_ROOT/mpi/sunhpc/loose-integration/suspend_sunmpi.sh "'$job_pid' \
  <queue_object_name>

4.3. PE object
--------------

The "qconf -spl" will show PE objects defined in the Grid Engine system. If
the administrator uses the default name to create a PE file for the loose
integration, there will be a PE object called "hpc". The contents can be
viewed using "qconf -sp <pe_name>" command. The loose integration package
also includes a PE template as an example for the loose integration. A PE
object can be created interactively by using the command, "qconf -ap
<pe_name>".


4.4. User Access List (ACL) object
----------------------------------

All the available list of ACL objects can be obtained using the "qconf -sul"
command. A new ACL object for HPC ClusterTools, called as HPCusers by
default, is created by the loose integration script. Administrator can
choose to use one from the existing ACL lists as well. The following shows
the format of the ACL.

	name       HPCusers
	entries    user1,user2,user3,...

The contents of an ACL can be viewed by using "qconf -su <acl_name>".  Also, 
the interactive command to create an ACL is "qconf -au <user1,user2,...> 
<acl_name>".
 
 

5. Limitations in the Current Loose Integration
-----------------------------------------------

- The suspend, resume and terminate scripts use the CRE Job Id of an MPI job. 
Currently these scripts pick up the Job Id by using adb utility with the
core file of an mprun process. The MPRUN wrapper should be used instead of
"mprun" command to store the process Id of an mprun process to a temporary
file so that both suspend and terminate scripts can use it. Without the Job
Id, since those scripts will not work properly, Grid Engine commands for
signalling an MPI job will not actually signal the MPI job. Therefore, it
breaks the loose integration.

- The "mprun" command should use a rank map file (-Mf or -m option depending
on ClusterTools' version) generated by Grid Engine. In this way, the MPI job
will be executed on resources allocated by Grid Engine. In some commercial
applications, mprun uses an explicit rank map file defined by its own
resource allocation scheme. Then, there may be a conflict between resources
allocated by Grid Engine and the application. However, as long as there is
enough resource available, the application will run and Grid Engine will
report bogus information about resource usage.

- The Grid Engine accounting information for an MPI job does not have 
much useful information except the wall-clock time. Also CPU time, memory
and IO usage information are not available with Sun Grid Engine.

- The resource control with Grid Engine works only with the wall-clock time.

- When an interactive job is started via a PE, if an MPI job is not started 
yet, then suspend, resume, and terminate methods will not work because there
is no CRE job Id.

- When the prism wrapper, PRISM, is used, suspend, resume, and terminate
methods will not work because those methods can not handle the fact that
Prism debugging job creates three distinct CRE Job Ids yet.
 

6. Test for the Loose Integration
---------------------------------

A user can submit an MPI job using either the interactive command (qrsh) or
the batch command (qsub) with a batch script. The following script is an
example batch script. A user can customize this file according to a
site-specific loose integration configuration and user information.

6.1. Batch script for HPC ClusterTools
--------------------------------------

The script

   batch.sunmpi.template

in this directory contains a template for a Grid Engine batch job.


6.2. Interactive command for HPC ClusterTools
---------------------------------------------

% qrsh <option list> -l <cre_exclusive_complex> \
  -pe <pe_name> <num_of_cpus> MPRUN -np <num_of_cpus>  \
  <mpi_executable>

Users can add more Grid Engine options to the option list such as those
shown in the batch example.


6.3. Prism debugging job
------------------------

% setenv DISPLAY `hostname`:0
% setenv PATH ${PATH}:/opt/SUNWhpc/bin
% qrsh  -V -l cre -pe hpc 4 PRISM -np 4 ./monte


7. Other Useful Information:

The official web site for Sun HPC ClusterTools, http://www.sun.com/hpc/ has
the latest product information in addition to many useful links.


Appendix A: Step by Step Loose Integration
------------------------------------------

The loose integration between Grid Engine and HPC ClusterTools can be done
by creating a parallel environment, exclusive complex and queues for HPC
ClusterTools, and appropriate suspend/terminate methods for the exclusive
queues associated with execution hosts running HPC ClusterTools.  By
creating an exclusive environment for HPC ClusterTools, the existing Grid
Engine installation will not be disturbed by this loose integration and vice
versa.  The following will show a step-by-step procedure for the loose
integration as a guideline.


Step 1: Create appropriate scripts for the loose integration
------------------------------------------------------------

The following scripts are essential to create the loose integration between
Sun Grid Engine and HPC ClusterTools. The current loose integration is
accomplished through Grid Engine interface and HPC ClusterTools commands,
mpinfo, mpps, and mpkill. The startsunmpi.sh is the most important script to
run an MPI job of HPC ClusterTools on Grid Engine system. The other scripts,
MPRUN, suspend_sunmpi.sh, terminate_sunmpi.sh, stopsunmpi.sh, are used to
provide robust support for control of an MPI job through the Grid Engine
commands. Create the following files and place them in $SGE_ROOT/mpi/sunhpc/
loose-integration directory, where $SGE_ROOT is an absolute path for the
Grid Engine installation directory.


A.1. MPRUN: a wrapper for the mprun command of HPC ClusterTools
---------------------------------------------------------------

The script

   MPRUN

in this directory contains a wrapper for the mprun command of HPC
ClusterTools.


A.2. PRISM: a wrapper for the prism command of HPC ClusterTools
---------------------------------------------------------------

The script

   PRISM

in this directory contains a wrapper for the prism command of HPC
ClusterTools.


A.3. startsunmpi.sh: a script to prepare a rank map file for HPC ClusterTools
-----------------------------------------------------------------------------

The script

   startsunmpi.sh

in this directory is a template script to prepare a rank map file.


A.4. stopsunmpi.sh: a script to be executed at the end of MPI jobs
------------------------------------------------------------------

The script

   stopsunmpi.sh

in this directory is executed at the end of MPI jobs.


A.5. suspend_sunmpi.sh: a script to suspend MPI jobs 
----------------------------------------------------

The script

   suspend_sunmpi.sh

in this directory suspends MPI jobs.


A.6. resume_sunmpi.sh: a script to resume a suspended MPI job
-------------------------------------------------------------

The script

   resume_sunmpi.sh

in this directory unsuspends MPI jobs. It uses the CRE Job Id saved by
suspend_sunmpi.sh script.


A.7. terminate_sunmpi.sh: a script to terminate MPI jobs
--------------------------------------------------------

The script

   terminate_sunmpi.sh

kin this directory kills MPI jobs. It will save all the process Ids spawned
by an MPI jobs.


Step 2: Create an exclusive complex for HPC ClusterTools
--------------------------------------------------------

In order to create a complex, which is a named container for a set of
attributes in Grid Engine, one can use "qconf -Ac" command or "qmon"
graphical user interface (GUI). The qconf command needs an input file
containing all the attributes for the exclusive complex. The following
complex definition is suggested for an exclusive complex for HPC
ClusterTools.

#name  shortcut  type  value  relop  requestable  consumable  default
#--------------------------------------------------------------------
cre    cre       BOOL  true   ==     FORCED       NO          false

Once this file is saved and named as "def_complex", we can create an
exclusive complex by issuing "qconf -Ac cre_exclusive def_complex". Here, it
is assumed that the exclusive complex is named as "cre_exclusive". If the
graphical user interface, qmon, is used, one can add a complex by clicking
the following buttons, "complex_configuration" -> "Add" -> (type attributes)
-> "Save" -> "Ok".


Step 3: Create exclusive queues for Sun HPC ClusterTools
--------------------------------------------------------

The exclusive queues for Sun HPC ClusterTools can be created either using
"qconf -Aq" with a queue definition file or using "qconf -aq" to retrieve
the default queue and modifying it accordingly, in addition to using the GUI
"qmon". It is recommended to create an exclusive queue for each execution
host running Sun HPC ClusterTools and name each exclusive queue as
<hostname>.cre. For example, one can create hodori.cre for an execution host
hodori. One needs to set the following lines of the queue attributes to make
it exclusive for Sun HPC ClusterTools.

qname <hostname>.cre
qtype BATCH  INTERACTIVE  PARALLEL
suspend_method $SGE_ROOT/mpi/sunhpc/loose-integration/suspend_sunmpi.sh $job_pid
resume_method $SGE_ROOT/mpi/sunhpc/loose-integration/resume_sunmpi.sh $job_pid
terminate_method $SGE_ROOT/mpi/sunhpc/loose-integration/terminate_sunmpi.sh $job_pid
complex_list cre_exclusive
complex_values cre=true

With the GUI qmon, select "Queue Control" button and click "Add" button to
set up the above information.


Step 4: Create a parallel environment for HPC ClusterTools
----------------------------------------------------------

In order to run an MPI job, the Grid Engine system requires a parallel
environment (PE). The following template file shows an example of a PE for
Sun HPC ClusterTools .

pe_name       <hpc>
queue_list    <list_of_exclusive_cre_queues>
slots         <number_of_cpus_available>
user_lists       <user_access_lists>
xuser_lists      NONE
start_proc_args  $SGE_ROOT/mpi/sunhpc/loose-integration/startsunmpi.sh $pe_hostfile /opt/SUNWhpc/bin
stop_proc_args   $SGE_ROOT/mpi/sunhpc/loose-integration/stopsunmpi.sh
allocation_rule  $fill_up
control_slaves    FALSE
job_is_first_task FALSE

All the fields surrounded by <> and $SGE_ROOT variable should be customized
or modified according to a site-specific information to create a PE. One can
interactively create a PE by using "qconf -ap <pe_name>" command to open an
edit window and filling out attributes following the above template. Also,
the GUI qmon can be used to create a parallel environment.  Select "Parallel
Environment Configuration" from the qmon main control window and click "Add"
button to bring up a window to fill up the above information.

The Grid Engine administrator can either create an exclusive user access
list (ACL) for Sun HPC ClusterTools or use an existing ACL for the parallel
environment setup. In order to create an ACL, the administrator can choose
one command among "qconf -Au", "qconf -au", and "qmon".


Copyright
---------

   (c) 2002 Sun Microsystems, Inc. Use is subject to license terms.
