DOKK / manpages / debian 11 / lam-runtime / procschema.5.en
PROCSCHEMA(5) LAM FILE FORMATS PROCSCHEMA(5)

procschema - LAM process schema format

#
# comment
#
<program> [$delay] [$inet_topo] [<arguments>]
<program> [$delay] [$inet_topo] [<arguments>]

...

Most LAM/MPI users can disregard this page.

A process schema (once ambiguously called a configuration file) lists the programs that will constitute the LAM environment on a particular node. It drives the operation of hboot(1). LAM system developers will find process schemata very useful for debugging and for generating custom systems. By convention these files begin with the prefix conf.

The default process schema selected by lamboot(1) (conf.lam) contains only one program, the LAM daemon (lamd). A one program process schema makes the whole business of process schemata and hboot(1) rather redundant. LAM can also be run in a de-clustered mode with the daemon reduced to a simple local message-passing server (the "kernel") and several system clients for network message-passing and remote services. This form of LAM is described in the process schema, conf.otb.

The syntax is line oriented. Comments begin with # and terminate with a newline.

Process lines consist of a filename, command line arguments, and possibly options and substitution variables. The command line arguments are passed to the process when it is started. The process options control how the process is started. Currently supported process options are:

$delay
After starting the process, pause before starting the next process.

Substitution variables are set by the tools that interpret the process schema and are a way of customizing the process at runtime. See hboot(1). Currently supported substitution variables are:

$inet_topo
typically, command-line arguments for LAM Internet datalink processes
$rtr_topo
typically, command-line arguments for the LAM network information process

The programs found in the LAM de-clustered mode process schema, conf.otb, are listed below.

bufferd	Creates, kills, sweeps, and states buffers.
bforward	Forward messages; helper for bufferd.
died	Monitors for death of user processes.
dli_inet	UDP/IP incoming connection to other nodes
dlo_inet	UDP/IP outgoing connection to other nodes
echod	Echoes messages; can be used to test nodes and links.
filed	Serves file access.
flatd	Provides symbolic access to node memory.
kenyad	Controls and monitors processes.
kernel	Coordinates local message-passing.
loadd	Loads executable files onto nodes.
router	Maintains network information.
traced	Collects and transports trace data.

Processes are started in the order given in the process schema, and for LAM, the order is important. In particular, the kernel must be first.

The de-clustered debug mode LAM process schema is shown below:

##
## The kernel is listed first.
##
kernel $delay
##
## daemons
##
router
kenyad
dli_inet $inet_topo
dlo_inet
bufferd
bforward
loadd
echod
flatd
filed
traced

$LAMHOME/etc/lam-conf.lamd
default process schema for lamboot(1), where $LAMHOME is the installation directory
$LAMHOME/etc/lam-conf.separate
default process schema for hboot(1)

lamboot(1), hboot(1)

July, 2007 LAM 7.1.4