PRM man pages : jobmngr(1)
NAME
jobmngr - Execute a PRM job
SYNOPSIS
jobmngr [ -a ] [ -Dlev ] [ -d displayname ] [ -h ] [ -m ] [
-nnice-value ] [ -r ] [ -s shost ] [ -j cfile ] [ -i task-
arguments ]
DESCRIPTION
jobmngr runs applications under PRM. A PRM job consists of
one or more application-tasks and a set of server tasks for
handling terminal and file I/O and collective communication
between user-tasks. Job configuration information is read
from a file specified on the command line. See USAGE below
for the format of the configuration file.
jobmngr sequentially requests nodes from system managers
specified on the command line and in the configuration file,
stopping when its resource requirements are satisfied. It
initiates application tasks on the assigned nodes. During
job execution, it responds to requests for spawning addi-
tional tasks, mapping task identifiers to host addresses and
migration of tasks. After all application tasks terminate,
jobmngr kills the server tasks and itself terminates.
When running, jobmngr writes status information to a status
file, .prm_job.jid in the user's home directory. This infor-
mation is used by utility programs such as prm_jobs(1) to
check on job status. The status file is deleted upon job
completion.
OPTIONS
-a Asynchronous task startup. By default, jobmngr sends a
synchronizing message to the tasks as part of their
initialization. With the -a option, no message is sent
after tasks are spawned.
-Dlev
Debug level. Print informational and debugging mes-
sages. More information is printed at larger values of
lev. -D1 and -D are equivalent.
-d displayname
Open an xterm(1) window on display displayname, to
which terminal output of the tasks is directed. This
only works a when running withing valid X window
environment. If creation of an xterm fails due to some
reason, the jobmngr will terminate.
-h Hold resources until job completes. Unless this option
is specified a node acquired from a system manager is
released immediately after tasks on the node complete.
-i task-options
Pass subsequent arguments to all tasks. Parsing of
jobmngr options stops when a -i is encountered.
-j jcfile
Read job configuration from file jcfile.
-m Acquire as many nodes as possible from the system
manager. If the number of tasks specified in the confi-
guration file is lesser than the number of nodes
acquired, tasks will be replicated.
-nniceval
Spawn tasks with reduced priority. Requests node
managers to nice all tasks to value niceval.
-r Activate replay debugging on those tasks for which one
of the 'r' options are specified in the job configura-
tion file. If this option is not specified, all 'r'
options (i.e. debugging, tracing, logging and replay)
in the configuration file are ignored.
-s shost
Prepend host shost to the list of system manager hosts
that jobmngr will attempt to acquire resources from.
This option can be repeated more than once.
USAGE
Job Configuration
A job configuration file provides a description of the job
to the job manager. Every PRM job requires a configuration
file. In this file, job configuration information is speci-
fied as attribute-value pairs and takes the form:
attribute = value
attribute = value
....
The attributes and the values they may take are listed
below:
Attr. Type Default Description
Value
t string none File name of task executable.
a string none Command line argument to task.
c boolean true This attribute is ignored if PRM has not
been configured and installed with the
'ckpt' option. If true, task executable
has been linked with Condor's libraries
and is capable of checkpointing and
migration.
d string $PRM_TOP/bin Parent directory name of the directory in
which the executable file resides. For
example: If the executable file 'ringsim'
resides in $PRM_TOP/bin/sun4, the value
of this attribute must be set to
$PRM_TOP/bin. PRM assumes that all exe-
cutables reside in a directory whose name
is the same as the architecture name of
the host machine. For example, on a sparc
machine, PRM expects executables to be in
a directory "sun4". In specifying a value
for attribute, this last component of the
directory name must be omitted.
f boolean false If true, a file_iotask is started by the
job manager during job initialization. If
false, and the program does require
remote file access, file_iotask will be
started by the job manager when a task
first tries to perform a remote-file-I/O
operation.
g boolean false If true, a group server (pvmgs) is
started by the job manager during job
initialization. If false, and a task
tries to perform a collective communica-
tion operation, the group server will be
started by the job manager.
i boolean false If true, terminal input will be
redirected to the task currently being
described.
n numeric 1 number of copies of the task to start up.
p boolean true If true, executable has been linked with
PRM's libraries.
rd boolean false This attribute is ignored if PRM has not
been configured and installed with the
'debug' option. If true, and jobmngr was
invoked with the -r option, task is
started under a debugger.
rf string none This attribute is ignored if not PRM has
been configured and installed with the
'debug' option. If jobmngr was invoked
with the -r option, the specified value
is taken as the name of file to which
trace data is written to or read from.
rl boolean false This attribute is ignored if PRM has not
been configured and installed with the
'debug' option. If true, and jobmngr was
invoked with the -r option, contents of
messages exchanged by the task are
logged.
rp boolean false This attribute is ignored if PRM has not
been configured and installed with the
debug option. If true, and a trace file
is specified with the rf attribute, and
jobmngr was invoked with the -r option,
replay is activated.
rt boolean false This attribute is ignored if PRM has not
been configured and installed with the
debug option. If true, and jobmngr was
invoked with the -r option, message trac-
ing is enabled.
s string provided in System manager hostname. More than one s
prm_config.h attribute may be specified.
Notes:
1) Any of these attributes except for t, may be left
unspecified. jobmngr will assume the default values.
2) System manager hosts specified using the -s option on
the command line are queried before those listed in the
configuration file.
3) Values for a can also be specified as command line
arguments to the job manager using the -i option. Argu-
ments specified this way are are passed on to all
tasks.
4) If the value for a d attribute starts with "io:"
(without quotes) task executable is taken as resident
on a remote filesystem that is not accessible to the
node on which the task is intended to run. The node
manager will then co-operate with file_iotask in
transferring the file to its local filesystem.
ENVIRONMENT VARIABLES
PRM_TOP
Top level of the directory hierarchy in which PRM is
installed. If this is not set, jobmngr determines the
PRM binaries directory using the value specified during
configuration.
DIAGNOSTICS
If any errors occur during initialization, acquisition of
nodes or creation of application/server tasks, jobmngr
prints a error message to stderr and exits after killing all
tasks and releasing nodes. If an application task dies due
to an exception or other reasons, a message is simply
printed, but the job itself is not terminated until all
tasks are done.
FILES
$HOME/.prm_job.pid job status file
SEE ALSO
nodemngr(8), prm_jobs(1), sysmngr(8)