6166278e18
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.
120 строки
4.7 KiB
C
120 строки
4.7 KiB
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:
|
|
*
|
|
* Populates global structure with process-specific information.
|
|
*
|
|
*
|
|
*/
|
|
|
|
#ifndef _ORTE_PROC_INFO_H_
|
|
#define _ORTE_PROC_INFO_H_
|
|
|
|
#include "orte_config.h"
|
|
#include "orte/types.h"
|
|
|
|
#ifdef HAVE_SYS_TYPES_H
|
|
#include <sys/types.h>
|
|
#endif
|
|
|
|
BEGIN_C_DECLS
|
|
|
|
/**
|
|
* Process information structure
|
|
*
|
|
* The orte_proc_info() function fills the pid field and obtains the
|
|
* process name, storing that information in the global structure. The
|
|
* structure also holds path names to the universe, job, and process
|
|
* session directories, and to the stdin, stdout, and stderr temp
|
|
* files - however, these are all initialized elsewhere.
|
|
*/
|
|
struct orte_proc_info_t {
|
|
orte_process_name_t my_name; /**< My official process name */
|
|
orte_process_name_t my_daemon; /**< Name of my local daemon */
|
|
char *my_daemon_uri; /**< Contact info to local daemon */
|
|
orte_std_cntr_t num_daemons; /**< number of active daemons */
|
|
orte_process_name_t my_hnp; /**< Name of my hnp */
|
|
char *my_hnp_uri; /**< Contact info for my hnp */
|
|
pid_t hnp_pid; /**< hnp pid - used if singleton */
|
|
orte_std_cntr_t app_num; /**< our index into the app_context array */
|
|
orte_std_cntr_t universe_size; /**< the size of the universe we are in */
|
|
orte_vpid_t num_procs; /**< number of processes in this job */
|
|
orte_vpid_t local_rank; /**< local rank on this node */
|
|
orte_std_cntr_t num_local_procs; /**< total number of procs on this node */
|
|
orte_nodeid_t nodeid; /**< numerical id for this node */
|
|
char *nodename; /**< string name for this node */
|
|
pid_t pid; /**< Local process ID for this process */
|
|
bool singleton; /**< I am a singleton */
|
|
bool daemon; /**< Indicate whether or not I am a daemon */
|
|
bool hnp; /**< Indicate whether or not I am the HNP (orterun) */
|
|
bool tool; /**< I am a tool or not */
|
|
/* The session directory has the form
|
|
* <prefix>/<openmpi-sessions-user>/<jobid>/<procid>, where the prefix
|
|
* can either be provided by the user via the
|
|
* --tmpdir command-line flag, the use of one of several
|
|
* environmental variables, or else a default location.
|
|
*/
|
|
char *tmpdir_base; /**< Base directory of the session dir tree */
|
|
char *top_session_dir; /**< Top-most directory of the session tree */
|
|
char *job_session_dir; /**< Session directory for job */
|
|
char *proc_session_dir; /**< Session directory for the process */
|
|
|
|
char *sock_stdin; /**< Path name to temp file for stdin. */
|
|
char *sock_stdout; /**< Path name to temp file for stdout. */
|
|
char *sock_stderr; /**< Path name to temp file for stderr. */
|
|
};
|
|
typedef struct orte_proc_info_t orte_proc_info_t;
|
|
|
|
|
|
/**
|
|
*
|
|
* Global process info descriptor. Initialized to almost no
|
|
* meaningful information - data is provided by calling \c
|
|
* orte_rte_init() (which calls \c orte_proc_info() to fill in the
|
|
* structure).
|
|
*
|
|
* The exception to this rule is the \c orte_process_info.seed field,
|
|
* which will be initialized to \c false, but should be set to \c true
|
|
* before calling \c orte_rte_info() if the caller is a seed daemon.
|
|
*/
|
|
ORTE_DECLSPEC extern orte_proc_info_t orte_process_info;
|
|
|
|
|
|
/**
|
|
* \internal
|
|
*
|
|
* Global structure to store a wide range of information about the
|
|
* process. orte_proc_info populates a global variable with
|
|
* information about the process being executing. This function should
|
|
* be called only once, from orte_rte_init().
|
|
*
|
|
* @param None.
|
|
*
|
|
* @retval ORTE_SUCCESS Successfully initialized the various fields.
|
|
* @retval OMPI_ERROR Failed to initialize one or more fields.
|
|
*/
|
|
|
|
ORTE_DECLSPEC int orte_proc_info(void);
|
|
|
|
ORTE_DECLSPEC int orte_proc_info_finalize(void);
|
|
|
|
END_C_DECLS
|
|
|
|
#endif
|