2005-03-14 20:57:21 +00:00
/*
2005-11-05 19:57:48 +00:00
* Copyright ( c ) 2004 - 2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation . All rights reserved .
2011-06-23 20:38:02 +00:00
* Copyright ( c ) 2004 - 2011 The University of Tennessee and The University
2005-11-05 19:57:48 +00:00
* of Tennessee Research Foundation . All rights
* reserved .
2005-03-14 20:57:21 +00:00
* Copyright ( c ) 2004 - 2005 High Performance Computing Center Stuttgart ,
* University of Stuttgart . All rights reserved .
2005-03-24 12:43:37 +00:00
* Copyright ( c ) 2004 - 2005 The Regents of the University of California .
* All rights reserved .
2011-11-03 14:22:07 +00:00
* Copyright ( c ) 2006 - 2011 Cisco Systems , Inc . All rights reserved .
2012-04-06 14:23:13 +00:00
* Copyright ( c ) 2011 Los Alamos National Security , LLC .
* All rights reserved .
2005-03-14 20:57:21 +00:00
* $ COPYRIGHT $
*
* Additional copyrights may follow
*
* $ HEADER $
*/
# include "orte_config.h"
2008-02-28 01:57:57 +00:00
# include "orte/constants.h"
2005-03-14 20:57:21 +00:00
2009-04-29 00:49:23 +00:00
# ifdef HAVE_STRING_H
# include <string.h>
# endif
2008-06-18 03:15:56 +00:00
# if !ORTE_DISABLE_FULL_SUPPORT
2006-02-12 01:33:29 +00:00
# include "opal/mca/mca.h"
2009-02-14 02:26:12 +00:00
# include "opal/util/output.h"
2006-02-12 01:33:29 +00:00
# include "opal/mca/base/base.h"
# include "opal/mca/base/mca_base_param.h"
2005-03-14 20:57:21 +00:00
2009-08-11 02:51:27 +00:00
# include "orte/runtime/orte_globals.h"
2009-10-13 04:19:07 +00:00
# include "orte/util/show_help.h"
2006-10-03 17:40:00 +00:00
2006-09-14 21:29:51 +00:00
# include "orte/mca/rmaps/base/rmaps_private.h"
2008-06-18 03:15:56 +00:00
# endif
2006-02-12 01:33:29 +00:00
# include "orte/mca/rmaps/base/base.h"
2005-03-14 20:57:21 +00:00
/*
* The following file was created by configure . It contains extern
* statements and the definition of an array of pointers to each
* component ' s public mca_base_component_t struct .
*/
2005-07-04 18:24:58 +00:00
# include "orte/mca/rmaps/base/static-components.h"
2005-03-14 20:57:21 +00:00
2008-06-18 03:15:56 +00:00
# if ORTE_DISABLE_FULL_SUPPORT
/* have to include a bogus function here so that
* the build system sees at least one function
* in the library
*/
int orte_rmaps_base_open ( void )
{
return ORTE_SUCCESS ;
}
# else
2005-03-14 20:57:21 +00:00
/*
2006-09-14 21:29:51 +00:00
* Global variables
2005-03-14 20:57:21 +00:00
*/
2006-09-14 21:29:51 +00:00
orte_rmaps_base_t orte_rmaps_base ;
2005-03-14 20:57:21 +00:00
/**
* Function for finding and opening either all MCA components , or the one
* that was specifically requested via a MCA parameter .
*/
int orte_rmaps_base_open ( void )
{
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
int param , value , i ;
2006-10-07 15:45:24 +00:00
char * policy ;
2008-09-23 15:46:34 +00:00
bool btmp ;
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
orte_mapping_policy_t tmp = 0 ;
orte_ranking_policy_t rtmp = 0 ;
char * * ck , * * ck2 ;
size_t len ;
2005-03-14 20:57:21 +00:00
2008-02-28 01:57:57 +00:00
/* init the globals */
2011-02-15 23:24:31 +00:00
OBJ_CONSTRUCT ( & orte_rmaps_base . selected_modules , opal_list_t ) ;
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
orte_rmaps_base . ppr = NULL ;
orte_rmaps_base . cpus_per_rank = 1 ;
orte_rmaps_base . display_map = false ;
orte_rmaps_base . slot_list = NULL ;
orte_rmaps_base . mapping = 0 ;
orte_rmaps_base . ranking = 0 ;
2011-02-15 23:24:31 +00:00
2008-02-28 01:57:57 +00:00
/* Debugging / verbose output. Always have stream open, with
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
verbose set by the mca open system . . . */
2008-06-09 14:53:58 +00:00
orte_rmaps_base . rmaps_output = opal_output_open ( NULL ) ;
2005-03-14 20:57:21 +00:00
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
/* define default mapping policy */
param = mca_base_param_reg_string_name ( " rmaps " , " base_mapping_policy " ,
# if OPAL_HAVE_HWLOC
" Mapping Policy [slot (default) | hwthread | core | l1cache | l2cache | l3cache | socket | numa | board | node | seq], with allowed modifiers :SPAN,OVERSUBSCRIBE,NOOVERSUBSCRIBE " ,
# else
" Mapping Policy [slot (default) | node], with allowed modifiers :SPAN,OVERSUBSCRIBE,NOOVERSUBSCRIBE " ,
# endif
false , false , NULL , & policy ) ;
mca_base_param_reg_syn_name ( param , " rmaps " , " base_schedule_policy " , true ) ;
if ( NULL = = policy ) {
ORTE_SET_MAPPING_POLICY ( orte_rmaps_base . mapping , ORTE_MAPPING_BYSLOT ) ;
ORTE_UNSET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_GIVEN ) ;
} else {
ck = opal_argv_split ( policy , ' : ' ) ;
if ( 2 < opal_argv_count ( ck ) ) {
/* incorrect format */
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-policy " , true , " mapping " , policy ) ;
opal_argv_free ( ck ) ;
return ORTE_ERR_SILENT ;
}
if ( 2 = = opal_argv_count ( ck ) ) {
ck2 = opal_argv_split ( ck [ 1 ] , ' , ' ) ;
for ( i = 0 ; NULL ! = ck2 [ i ] ; i + + ) {
if ( 0 = = strncasecmp ( ck2 [ i ] , " span " , strlen ( ck2 [ i ] ) ) ) {
orte_rmaps_base . mapping | = ORTE_MAPPING_SPAN ;
} else if ( 0 = = strncasecmp ( ck2 [ i ] , " oversubscribe " , strlen ( ck2 [ i ] ) ) ) {
if ( ORTE_MAPPING_SUBSCRIBE_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) {
/* error - cannot redefine the default mapping policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" oversubscribe " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_UNSET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_NO_OVERSUBSCRIBE ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_SUBSCRIBE_GIVEN ) ;
} else if ( 0 = = strncasecmp ( ck2 [ i ] , " nooversubscribe " , strlen ( ck2 [ i ] ) ) ) {
if ( ORTE_MAPPING_SUBSCRIBE_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) {
/* error - cannot redefine the default mapping policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" nooversubscribe " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_NO_OVERSUBSCRIBE ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_SUBSCRIBE_GIVEN ) ;
} else {
/* unrecognized modifier */
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-modifier " , true , " mapping " , ck2 [ i ] ) ;
opal_argv_free ( ck ) ;
opal_argv_free ( ck2 ) ;
return ORTE_ERR_SILENT ;
}
}
opal_argv_free ( ck2 ) ;
}
len = strlen ( ck [ 0 ] ) ;
if ( 0 = = strncasecmp ( ck [ 0 ] , " slot " , len ) ) {
tmp = ORTE_MAPPING_BYSLOT ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " node " , len ) ) {
tmp = ORTE_MAPPING_BYNODE ;
# if OPAL_HAVE_HWLOC
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " core " , len ) ) {
tmp = ORTE_MAPPING_BYCORE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l1cache " , len ) ) {
tmp = ORTE_MAPPING_BYL1CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l2cache " , len ) ) {
tmp = ORTE_MAPPING_BYL2CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l3cache " , len ) ) {
tmp = ORTE_MAPPING_BYL3CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " socket " , len ) ) {
tmp = ORTE_MAPPING_BYSOCKET ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " numa " , len ) ) {
tmp = ORTE_MAPPING_BYNUMA ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " board " , len ) ) {
tmp = ORTE_MAPPING_BYBOARD ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " hwthread " , len ) ) {
tmp = ORTE_MAPPING_BYHWTHREAD ;
/* if we are mapping processes to individual hwthreads, then
* we need to treat those hwthreads as separate cpus
*/
opal_hwloc_use_hwthreads_as_cpus = true ;
# endif
} else {
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-policy " , true , " mapping " , policy ) ;
opal_argv_free ( ck ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_MAPPING_POLICY ( orte_rmaps_base . mapping , tmp ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_GIVEN ) ;
opal_argv_free ( ck ) ;
}
/* define default ranking policy */
param = mca_base_param_reg_string_name ( " rmaps " , " base_ranking_policy " ,
# if OPAL_HAVE_HWLOC
" Ranking Policy [slot (default) | hwthread | core | l1cache | l2cache | l3cache | socket | numa | board | node], with modifier :SPAN or :FILL " ,
# else
" Ranking Policy [slot (default) | node] " ,
# endif
false , false , NULL , & policy ) ;
if ( NULL = = policy ) {
ORTE_SET_RANKING_POLICY ( orte_rmaps_base . ranking , ORTE_RANK_BY_SLOT ) ;
} else {
ck = opal_argv_split ( policy , ' : ' ) ;
if ( 2 < opal_argv_count ( ck ) ) {
/* incorrect format */
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-policy " , true , " ranking " , policy ) ;
opal_argv_free ( ck ) ;
return ORTE_ERR_SILENT ;
}
if ( 2 = = opal_argv_count ( ck ) ) {
if ( 0 = = strncasecmp ( ck [ 1 ] , " span " , strlen ( ck [ 1 ] ) ) ) {
orte_rmaps_base . ranking | = ORTE_RANKING_SPAN ;
} else if ( 0 = = strncasecmp ( ck [ 1 ] , " fill " , strlen ( ck [ 1 ] ) ) ) {
orte_rmaps_base . ranking | = ORTE_RANKING_FILL ;
} else {
/* unrecognized modifier */
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-modifier " , true , " ranking " , ck [ 1 ] ) ;
opal_argv_free ( ck ) ;
return ORTE_ERR_SILENT ;
}
}
len = strlen ( ck [ 0 ] ) ;
if ( 0 = = strncasecmp ( ck [ 0 ] , " slot " , len ) ) {
rtmp = ORTE_RANK_BY_SLOT ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " node " , len ) ) {
rtmp = ORTE_RANK_BY_NODE ;
# if OPAL_HAVE_HWLOC
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " hwthread " , len ) ) {
rtmp = ORTE_RANK_BY_HWTHREAD ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " core " , len ) ) {
rtmp = ORTE_RANK_BY_CORE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l1cache " , len ) ) {
rtmp = ORTE_RANK_BY_L1CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l2cache " , len ) ) {
rtmp = ORTE_RANK_BY_L2CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " l3cache " , len ) ) {
rtmp = ORTE_RANK_BY_L3CACHE ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " socket " , len ) ) {
rtmp = ORTE_RANK_BY_SOCKET ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " numa " , len ) ) {
rtmp = ORTE_RANK_BY_NUMA ;
} else if ( 0 = = strncasecmp ( ck [ 0 ] , " board " , len ) ) {
rtmp = ORTE_RANK_BY_BOARD ;
# endif
} else {
orte_show_help ( " help-orte-rmaps-base.txt " , " unrecognized-policy " , true , " ranking " , policy ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_RANKING_POLICY ( orte_rmaps_base . ranking , rtmp ) ;
ORTE_SET_RANKING_DIRECTIVE ( orte_rmaps_base . ranking , ORTE_RANKING_GIVEN ) ;
}
/* backward compatibility */
mca_base_param_reg_int_name ( " rmaps " , " base_byslot " ,
" Whether to map and rank processes round-robin by slot " ,
false , false , ( int ) false , & value ) ;
if ( value ) {
/* set mapping policy to byslot - error if something else already set */
if ( ( ORTE_MAPPING_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) & &
ORTE_GET_MAPPING_POLICY ( orte_rmaps_base . mapping ) ! = ORTE_MAPPING_BYSLOT ) {
/* error - cannot redefine the default mapping policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" byslot " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_MAPPING_POLICY ( orte_rmaps_base . mapping , ORTE_MAPPING_BYSLOT ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_GIVEN ) ;
/* set ranking policy to byslot - error if something else already set */
if ( ( ORTE_RANKING_GIVEN & ORTE_GET_RANKING_DIRECTIVE ( orte_rmaps_base . ranking ) ) & &
ORTE_GET_RANKING_POLICY ( orte_rmaps_base . ranking ) ! = ORTE_RANK_BY_SLOT ) {
/* error - cannot redefine the default ranking policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " ranking " ,
" byslot " , orte_rmaps_base_print_ranking ( orte_rmaps_base . ranking ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_RANKING_POLICY ( orte_rmaps_base . ranking , ORTE_RANK_BY_SLOT ) ;
ORTE_SET_RANKING_DIRECTIVE ( orte_rmaps_base . ranking , ORTE_RANKING_GIVEN ) ;
}
mca_base_param_reg_int_name ( " rmaps " , " base_bynode " ,
" Whether to map and rank processes round-robin by node " ,
false , false , ( int ) false , & value ) ;
if ( value ) {
/* set mapping policy to bynode - error if something else already set */
if ( ( ORTE_MAPPING_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) & &
ORTE_GET_MAPPING_POLICY ( orte_rmaps_base . mapping ) ! = ORTE_MAPPING_BYNODE ) {
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" bynode " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_MAPPING_POLICY ( orte_rmaps_base . mapping , ORTE_MAPPING_BYNODE ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_GIVEN ) ;
/* set ranking policy to bynode - error if something else already set */
if ( ( ORTE_RANKING_GIVEN & ORTE_GET_RANKING_DIRECTIVE ( orte_rmaps_base . ranking ) ) & &
ORTE_GET_RANKING_POLICY ( orte_rmaps_base . ranking ) ! = ORTE_RANK_BY_NODE ) {
/* error - cannot redefine the default ranking policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " ranking " ,
" bynode " , orte_rmaps_base_print_ranking ( orte_rmaps_base . ranking ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_RANKING_POLICY ( orte_rmaps_base . ranking , ORTE_RANK_BY_NODE ) ;
ORTE_SET_RANKING_DIRECTIVE ( orte_rmaps_base . ranking , ORTE_RANKING_GIVEN ) ;
2005-10-07 22:24:52 +00:00
}
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
#if 0
2009-08-11 02:51:27 +00:00
/* #cpus/rank to use */
2009-10-13 04:19:07 +00:00
param = mca_base_param_reg_int_name ( " rmaps " , " base_cpus_per_proc " ,
2009-08-11 02:51:27 +00:00
" Number of cpus to use for each rank [1-2**15 (default=1)] " ,
2009-08-30 14:30:36 +00:00
false , false , 1 , NULL ) ;
mca_base_param_reg_syn_name ( param , " rmaps " , " base_cpus_per_rank " , false ) ;
mca_base_param_lookup_int ( param , & value ) ;
2009-08-11 02:51:27 +00:00
orte_rmaps_base . cpus_per_rank = value ;
2009-08-26 02:01:49 +00:00
/* if the cpus/rank > 1, then we have to bind to cores UNLESS the binding has
* already been set to something else
*/
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
if ( 1 < orte_rmaps_base . cpus_per_rank & &
! OPAL_BINDING_POLICY_IS_SET ( opal_hwloc_binding_policy ) ) {
opal_hwloc_binding_policy | = OPAL_BIND_TO_CORE ;
2009-08-11 02:51:27 +00:00
}
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
# endif
2006-07-04 20:12:35 +00:00
/* Should we schedule on the local node or not? */
2007-03-24 16:16:16 +00:00
mca_base_param_reg_int_name ( " rmaps " , " base_no_schedule_local " ,
" If false, allow scheduling MPI applications on the same node as mpirun (default). If true, do not schedule any MPI applications on the same node as mpirun " ,
false , false , ( int ) false , & value ) ;
2008-04-09 22:10:53 +00:00
if ( value ) {
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
orte_rmaps_base . mapping | = ORTE_MAPPING_NO_USE_LOCAL ;
2008-04-09 22:10:53 +00:00
}
2006-07-04 20:12:35 +00:00
2006-07-10 14:10:21 +00:00
/* Should we oversubscribe or not? */
2008-02-28 01:57:57 +00:00
/** default condition that allows oversubscription */
2006-07-10 14:10:21 +00:00
mca_base_param_reg_int_name ( " rmaps " , " base_no_oversubscribe " ,
2007-03-24 16:16:16 +00:00
" If true, then do not allow oversubscription of nodes - mpirun will return an error if there aren't enough nodes to launch all processes without oversubscribing " ,
false , false , ( int ) false , & value ) ;
2008-02-28 01:57:57 +00:00
if ( value ) {
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
if ( ( ORTE_MAPPING_SUBSCRIBE_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) & &
! ( ORTE_MAPPING_NO_OVERSUBSCRIBE & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) ) {
/* error - cannot redefine the default mapping policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" no-oversubscribe " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_NO_OVERSUBSCRIBE ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_SUBSCRIBE_GIVEN ) ;
2006-07-10 14:10:21 +00:00
}
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
/** force oversubscription permission */
mca_base_param_reg_int_name ( " rmaps " , " base_oversubscribe " ,
" If true, then =allow oversubscription of nodes " ,
false , false , ( int ) false , & value ) ;
if ( value ) {
if ( ( ORTE_MAPPING_SUBSCRIBE_GIVEN & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) & &
( ORTE_MAPPING_NO_OVERSUBSCRIBE & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) ) {
/* error - cannot redefine the default mapping policy */
orte_show_help ( " help-orte-rmaps-base.txt " , " redefining-policy " , true , " mapping " ,
" oversubscribe " , orte_rmaps_base_print_mapping ( orte_rmaps_base . mapping ) ) ;
return ORTE_ERR_SILENT ;
}
ORTE_UNSET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_NO_OVERSUBSCRIBE ) ;
ORTE_SET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping , ORTE_MAPPING_SUBSCRIBE_GIVEN ) ;
}
2006-11-17 19:06:10 +00:00
/* should we display the map after determining it? */
2007-08-18 03:05:45 +00:00
mca_base_param_reg_int_name ( " rmaps " , " base_display_map " ,
2006-11-17 19:06:10 +00:00
" Whether to display the process map after it is computed " ,
false , false , ( int ) false , & value ) ;
2008-02-28 01:57:57 +00:00
orte_rmaps_base . display_map = OPAL_INT_TO_BOOL ( value ) ;
2006-07-10 14:10:21 +00:00
2008-09-23 15:46:34 +00:00
/* should we display a detailed (developer-quality) version of the map after determining it? */
mca_base_param_reg_int_name ( " rmaps " , " base_display_devel_map " ,
" Whether to display a developer-detail process map after it is computed " ,
false , false , ( int ) false , & value ) ;
btmp = OPAL_INT_TO_BOOL ( value ) ;
if ( btmp ) {
orte_rmaps_base . display_map = true ;
orte_devel_level_output = true ;
}
2011-10-29 15:12:45 +00:00
/* should we display the topology along with the map? */
mca_base_param_reg_int_name ( " rmaps " , " base_display_topo_with_map " ,
" Whether to display the topology with the map " ,
false , false , ( int ) false , & value ) ;
orte_display_topo_with_map = OPAL_INT_TO_BOOL ( value ) ;
2011-11-03 14:22:07 +00:00
/* should we display a diffable report of proc locations after determining it? */
mca_base_param_reg_int_name ( " rmaps " , " base_display_diffable_map " ,
" Whether to display a diffable process map after it is computed " ,
false , false , ( int ) false , & value ) ;
btmp = OPAL_INT_TO_BOOL ( value ) ;
if ( btmp ) {
orte_rmaps_base . display_map = true ;
orte_display_diffable_output = true ;
}
2005-03-14 20:57:21 +00:00
/* Open up all the components that we can find */
if ( ORTE_SUCCESS ! =
2005-08-15 18:25:35 +00:00
mca_base_components_open ( " rmaps " , orte_rmaps_base . rmaps_output ,
mca_rmaps_base_static_components ,
2008-02-28 01:57:57 +00:00
& orte_rmaps_base . available_components , true ) ) {
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
return ORTE_ERROR ;
2005-03-14 20:57:21 +00:00
}
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
/* check to see if any component indicated a problem */
if ( ORTE_MAPPING_CONFLICTED & ORTE_GET_MAPPING_DIRECTIVE ( orte_rmaps_base . mapping ) ) {
/* the component would have already reported the error, so
* tell the rest of the chain to shut up
*/
return ORTE_ERR_SILENT ;
}
2005-03-14 20:57:21 +00:00
At long last, the fabled revision to the affinity system has arrived. A more detailed explanation of how this all works will be presented here:
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
2011-11-15 03:40:11 +00:00
/* All done */
2005-03-14 20:57:21 +00:00
return ORTE_SUCCESS ;
}
2008-06-18 03:15:56 +00:00
2011-02-15 23:24:31 +00:00
OBJ_CLASS_INSTANCE ( orte_rmaps_base_selected_module_t ,
opal_list_item_t ,
NULL , NULL ) ;
2008-06-18 03:15:56 +00:00
# endif /* ORTE_DISABLE_FULL_SUPPORT */
