1
1
openmpi/orte/mca/odls/odls.h
Ralph Castain 6166278e18 Improve the scalability of the modex operation and fix a bug reported by Tim P
The bug was a race condition in the barrier operation that caused the barrier in MPI_Finalize to fail on very short programs.

Scalaiblity was improved by using the daemons to aggregate modex and barrier messages before sending them to the rank=0 proc. Improvement is proportional to ppn, of course, but there really wasn't a scaling problem at low ppn anyway. This modification also paves the way for better allgather operations since now all the data for each node is sitting at the daemon level, and the daemons are now aware that a collective operation on the OOB is underway (so they -can- participate in a collective of their own to support it).

Also added better diagnostics to map out the timing associated with MPI_Init - turned on by -mca orte_timing 1.

This commit was SVN r17988.
2008-03-27 15:17:53 +00:00

164 строки
5.2 KiB
C

/* -*- C -*-
*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* The OpenRTE Daemon's Local Launch Subsystem
*
*/
#ifndef ORTE_MCA_ODLS_H
#define ORTE_MCA_ODLS_H
#include "orte_config.h"
#include "orte/types.h"
#include "opal/mca/mca.h"
#include "opal/class/opal_list.h"
#include "opal/dss/dss_types.h"
#include "orte/mca/rmaps/rmaps_types.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/odls/odls_types.h"
BEGIN_C_DECLS
/*
* odls module functions
*/
/*
* Construct a buffer for use in adding local processes
* In order to reuse daemons, we need a way for the HNP to construct a buffer that
* contains the data needed by the active ODLS component to launch a local process. Since the
* only one that knows what a particular ODLS component needs is that component, we require an
* entry point that the HNP can call to get the required buffer. This is constructed
* for *all* nodes - the individual orteds then parse that data to find the specific launch info
* for procs on their node
*/
typedef int (*orte_odls_base_module_get_add_procs_data_fn_t)(opal_buffer_t *data,
orte_jobid_t job);
/**
* Locally launch the provided processes
*/
typedef int (*orte_odls_base_module_launch_local_processes_fn_t)(opal_buffer_t *data);
/**
* Kill the local processes on this node
*/
typedef int (*orte_odls_base_module_kill_local_processes_fn_t)(orte_jobid_t job, bool set_state);
/**
* Signal local processes
*/
typedef int (*orte_odls_base_module_signal_local_process_fn_t)(const orte_process_name_t *proc,
int32_t signal);
/**
* Deliver a message to local processes
*/
typedef int (*orte_odls_base_module_deliver_message_fn_t)(orte_jobid_t job, opal_buffer_t *buffer,
orte_rml_tag_t tag);
/**
* Register to require sync before termination
*/
typedef int (*orte_odls_base_module_require_sync_fn_t)(orte_process_name_t *proc, opal_buffer_t *buffer);
/**
* Collect data as part of a collective operation by the procs
*/
typedef int (*orte_odls_base_module_collect_data_fn_t)(orte_process_name_t *proc, opal_buffer_t *buffer);
/**
* pls module version 1.3.0
*/
struct orte_odls_base_module_1_3_0_t {
orte_odls_base_module_get_add_procs_data_fn_t get_add_procs_data;
orte_odls_base_module_launch_local_processes_fn_t launch_local_procs;
orte_odls_base_module_kill_local_processes_fn_t kill_local_procs;
orte_odls_base_module_signal_local_process_fn_t signal_local_procs;
orte_odls_base_module_deliver_message_fn_t deliver_message;
orte_odls_base_module_require_sync_fn_t require_sync;
orte_odls_base_module_collect_data_fn_t collect_data;
};
/** shorten orte_odls_base_module_1_3_0_t declaration */
typedef struct orte_odls_base_module_1_3_0_t orte_odls_base_module_1_3_0_t;
/** shorten orte_odls_base_module_t declaration */
typedef struct orte_odls_base_module_1_3_0_t orte_odls_base_module_t;
/**
* odls initialization function
*
* Called by the MCA framework to initialize the component. Invoked
* exactly once per process.
*
* @param priority (OUT) Relative priority or ranking use by MCA to
* select a module.
*/
typedef struct orte_odls_base_module_1_3_0_t*
(*orte_odls_base_component_init_fn_t)(int *priority);
/**
* Cleanup all resources held by the component
*/
typedef int (*orte_odls_base_component_finalize_fn_t)(void);
/**
* odls component v1.3.0
*/
struct orte_odls_base_component_1_3_0_t {
/** component version */
mca_base_component_t version;
/** component data */
mca_base_component_data_1_0_0_t odls_data;
/** Function called when component is initialized */
orte_odls_base_component_init_fn_t init;
/* Function called when component is finalized */
orte_odls_base_component_finalize_fn_t finalize;
};
/** Convenience typedef */
typedef struct orte_odls_base_component_1_3_0_t orte_odls_base_component_1_3_0_t;
/** Convenience typedef */
typedef orte_odls_base_component_1_3_0_t orte_odls_base_component_t;
/**
* Macro for use in modules that are of type odls v1.3.0
*/
#define ORTE_ODLS_BASE_VERSION_1_3_0 \
/* odls v1.3 is chained to MCA v1.0 */ \
MCA_BASE_VERSION_1_0_0, \
/* odls v1.3 */ \
"odls", 1, 3, 0
/* Global structure for accessing ODLS functions
*/
ORTE_DECLSPEC extern orte_odls_base_module_t orte_odls; /* holds selected module's function pointers */
END_C_DECLS
#endif /* MCA_ODLS_H */