54b2cf747e
The commit has been tested for C/R and Cray operations, and on Odin (SLURM, rsh) and RoadRunner (TM). I tried to update all environments, but obviously could not test them. I know that Windows needs some work, and have highlighted what is know to be needed in the odls process component. This represents a lot of work by Brian, Tim P, Josh, and myself, with much advice from Jeff and others. For posterity, I have appended a copy of the email describing the work that was done: As we have repeatedly noted, the modex operation in MPI_Init is the single greatest consumer of time during startup. To-date, we have executed that operation as an ORTE stage gate that held the process until a startup message containing all required modex (and OOB contact info - see #3 below) info could be sent to it. Each process would send its data to the HNP's registry, which assembled and sent the message when all processes had reported in. In addition, ORTE had taken responsibility for monitoring process status as it progressed through a series of "stage gates". The process reported its status at each gate, and ORTE would then send a "release" message once all procs had reported in. The incoming changes revamp these procedures in three ways: 1. eliminating the ORTE stage gate system and cleanly delineating responsibility between the OMPI and ORTE layers for MPI init/finalize. The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info. The allgather is implemented using the orte_grpcomm framework since the BTL's are not active at that point. At the moment, the grpcomm framework only has a "basic" component analogous to OMPI's "basic" coll framework - I would recommend that the MPI team create additional, more advanced components to improve performance of this step. The other stage gates have been replaced by orte_grpcomm barrier functions. We tried to use MPI barriers instead (since the BTL's are active at that point), but - as we discussed on the telecon - these are not currently true barriers so the job would hang when we fell through while messages were still in process. Note that the grpcomm barrier doesn't actually resolve that problem, but Brian has pointed out that we are unlikely to ever see it violated. Again, you might want to spend a little time on an advanced barrier algorithm as the one in "basic" is very simplistic. Summarizing this change: ORTE no longer tracks process state nor has direct responsibility for synchronizing jobs. This is now done via collective operations within the MPI layer, albeit using ORTE collective communication services. I -strongly- urge the MPI team to implement advanced collective algorithms to improve the performance of this critical procedure. 2. reducing the volume of data exchanged during modex. Data in the modex consisted of the process name, the name of the node where that process is located (expressed as a string), plus a string representation of all contact info. The nodename was required in order for the modex to determine if the process was local or not - in addition, some people like to have it to print pretty error messages when a connection failed. The size of this data has been reduced in three ways: (a) reducing the size of the process name itself. The process name consisted of two 32-bit fields for the jobid and vpid. This is far larger than any current system, or system likely to exist in the near future, can support. Accordingly, the default size of these fields has been reduced to 16-bits, which means you can have 32k procs in each of 32k jobs. Since the daemons must have a vpid, and we require one daemon/node, this also restricts the default configuration to 32k nodes. To support any future "mega-clusters", a configuration option --enable-jumbo-apps has been added. This option increases the jobid and vpid field sizes to 32-bits. Someday, if necessary, someone can add yet another option to increase them to 64-bits, I suppose. (b) replacing the string nodename with an integer nodeid. Since we have one daemon/node, the nodeid corresponds to the local daemon's vpid. This replaces an often lengthy string with only 2 (or at most 4) bytes, a substantial reduction. (c) when the mca param requesting that nodenames be sent to support pretty error messages, a second mca param is now used to request FQDN - otherwise, the domain name is stripped (by default) from the message to save space. If someone wants to combine those into a single param somehow (perhaps with an argument?), they are welcome to do so - I didn't want to alter what people are already using. While these may seem like small savings, they actually amount to a significant impact when aggregated across the entire modex operation. Since every proc must receive the modex data regardless of the collective used to send it, just reducing the size of the process name removes nearly 400MBytes of communication from a 32k proc job (admittedly, much of this comm may occur in parallel). So it does add up pretty quickly. 3. routing RML messages to reduce connections. The default messaging system remains point-to-point - i.e., each proc opens a socket to every proc it communicates with and sends its messages directly. A new option uses the orteds as routers - i.e., each proc only opens a single socket to its local orted. All messages are sent from the proc to the orted, which forwards the message to the orted on the node where the intended recipient proc is located - that orted then forwards the message to its local proc (the recipient). This greatly reduces the connection storm we have encountered during startup. It also has the benefit of removing the sharing of every proc's OOB contact with every other proc. The orted routing tables are populated during launch since every orted gets a map of where every proc is being placed. Each proc, therefore, only needs to know the contact info for its local daemon, which is passed in via the environment when the proc is fork/exec'd by the daemon. This alone removes ~50 bytes/process of communication that was in the current STG1 startup message - so for our 32k proc job, this saves us roughly 32k*50 = 1.6MBytes sent to 32k procs = 51GBytes of messaging. Note that you can use the new routing method by specifying -mca routed tree - if you so desire. This mode will become the default at some point in the future. There are a few minor additional changes in the commit that I'll just note in passing: * propagation of command line mca params to the orteds - fixes ticket #1073. See note there for details. * requiring of "finalize" prior to "exit" for MPI procs - fixes ticket #1144. See note there for details. * cleanup of some stale header files This commit was SVN r16364.
399 строки
15 KiB
C
399 строки
15 KiB
C
/*
|
|
* Copyright (c) 2007 Los Alamos National Security, LLC.
|
|
* All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "orte_config.h"
|
|
#include "orte/orte_constants.h"
|
|
|
|
#include "opal/util/output.h"
|
|
|
|
#include "orte/mca/errmgr/errmgr.h"
|
|
#include "orte/mca/ns/ns.h"
|
|
#include "orte/mca/rmaps/rmaps.h"
|
|
#include "orte/mca/grpcomm/grpcomm.h"
|
|
#include "orte/mca/odls/odls.h"
|
|
|
|
#include "orte/mca/smr/smr.h"
|
|
#include "orte/mca/rml/base/rml_contact.h"
|
|
|
|
#include "orte/mca/routed/base/base.h"
|
|
#include "routed_tree.h"
|
|
|
|
int
|
|
orte_routed_tree_update_route(orte_process_name_t *target,
|
|
orte_process_name_t *route)
|
|
{
|
|
if (target->jobid == ORTE_JOBID_INVALID ||
|
|
target->vpid == ORTE_VPID_INVALID) {
|
|
return ORTE_ERR_BAD_PARAM;
|
|
}
|
|
|
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
|
|
"%s routed_tree_update: %s --> %s",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
|
|
ORTE_NAME_PRINT(target),
|
|
ORTE_NAME_PRINT(route)));
|
|
|
|
|
|
/* exact match */
|
|
if (target->jobid != ORTE_JOBID_WILDCARD &&
|
|
target->vpid != ORTE_VPID_WILDCARD) {
|
|
opal_list_item_t *item;
|
|
orte_routed_tree_entry_t *entry;
|
|
|
|
for (item = opal_list_get_first(&orte_routed_tree_module.peer_list) ;
|
|
item != opal_list_get_end(&orte_routed_tree_module.peer_list) ;
|
|
item = opal_list_get_next(item)) {
|
|
entry = (orte_routed_tree_entry_t*) item;
|
|
|
|
if (0 == orte_ns.compare_fields(ORTE_NS_CMP_ALL,
|
|
target, &entry->target)) {
|
|
entry->route = *route;
|
|
return ORTE_SUCCESS;
|
|
}
|
|
}
|
|
|
|
entry = OBJ_NEW(orte_routed_tree_entry_t);
|
|
entry->target = *target;
|
|
entry->route = *route;
|
|
opal_list_append(&orte_routed_tree_module.peer_list, &entry->super);
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
/* vpid wildcard */
|
|
if (target->jobid != ORTE_JOBID_WILDCARD &&
|
|
target->vpid == ORTE_VPID_WILDCARD) {
|
|
opal_list_item_t *item;
|
|
orte_routed_tree_entry_t *entry;
|
|
|
|
for (item = opal_list_get_first(&orte_routed_tree_module.vpid_wildcard_list) ;
|
|
item != opal_list_get_end(&orte_routed_tree_module.vpid_wildcard_list) ;
|
|
item = opal_list_get_next(item)) {
|
|
entry = (orte_routed_tree_entry_t*) item;
|
|
|
|
if (0 == orte_ns.compare_fields(ORTE_NS_CMP_JOBID,
|
|
target, &entry->target)) {
|
|
entry->route = *route;
|
|
return ORTE_SUCCESS;
|
|
}
|
|
}
|
|
|
|
entry = OBJ_NEW(orte_routed_tree_entry_t);
|
|
entry->target = *target;
|
|
entry->route = *route;
|
|
opal_list_append(&orte_routed_tree_module.vpid_wildcard_list, &entry->super);
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
/* wildcard */
|
|
if (target->jobid == ORTE_JOBID_WILDCARD &&
|
|
target->vpid == ORTE_VPID_WILDCARD) {
|
|
orte_routed_tree_module.full_wildcard_entry.route = *route;
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
return ORTE_ERR_NOT_SUPPORTED;
|
|
}
|
|
|
|
|
|
orte_process_name_t
|
|
orte_routed_tree_get_route(orte_process_name_t *target)
|
|
{
|
|
orte_process_name_t ret;
|
|
opal_list_item_t *item;
|
|
|
|
/* if it is me, then the route is just direct */
|
|
if (ORTE_EQUAL == orte_dss.compare(ORTE_PROC_MY_NAME, target, ORTE_NAME)) {
|
|
ret = *target;
|
|
goto found;
|
|
}
|
|
|
|
/* check exact matches */
|
|
for (item = opal_list_get_first(&orte_routed_tree_module.peer_list) ;
|
|
item != opal_list_get_end(&orte_routed_tree_module.peer_list) ;
|
|
item = opal_list_get_next(item)) {
|
|
orte_routed_tree_entry_t *entry =
|
|
(orte_routed_tree_entry_t*) item;
|
|
|
|
if (0 == orte_ns.compare_fields(ORTE_NS_CMP_ALL,
|
|
target, &entry->target)) {
|
|
ret = entry->route;
|
|
goto found;
|
|
}
|
|
}
|
|
|
|
ret = orte_routed_tree_module.full_wildcard_entry.route;
|
|
|
|
found:
|
|
|
|
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
|
|
"%s routed_tree_get(%s) --> %s",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
|
|
ORTE_NAME_PRINT(target),
|
|
ORTE_NAME_PRINT(&ret)));
|
|
|
|
return ret;
|
|
}
|
|
|
|
int orte_routed_tree_init_routes(orte_jobid_t job, orte_gpr_notify_data_t *ndat)
|
|
{
|
|
/* the tree module routes all proc communications through
|
|
* the local daemon. Daemons must identify which of their
|
|
* daemon-peers is "hosting" the specified recipient and
|
|
* route the message to that daemon. Daemon contact info
|
|
* is handled elsewhere, so all we need to do here is
|
|
* ensure that the procs are told to route through their
|
|
* local daemon, and that daemons are told how to route
|
|
* for each proc
|
|
*/
|
|
int rc;
|
|
|
|
/* if I am a daemon or HNP, then I have to extract the routing info for this job
|
|
* from the data sent to me for launch and update the routing tables to
|
|
* point at the daemon for each proc
|
|
*/
|
|
if (orte_process_info.daemon || orte_process_info.seed) {
|
|
orte_std_cntr_t i, j;
|
|
orte_process_name_t daemon, proc;
|
|
orte_gpr_value_t **values, *value;
|
|
|
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
|
|
"%s routed_tree: init routes for daemon/seed job %ld",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)job));
|
|
|
|
if (0 == job) {
|
|
if (NULL == ndat) {
|
|
/* if ndat is NULL, then this is being called during init,
|
|
* so just seed the routing table with a path back to the HNP...
|
|
*/
|
|
if (ORTE_SUCCESS != (rc = orte_routed_tree_update_route(ORTE_PROC_MY_HNP,
|
|
ORTE_PROC_MY_HNP))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
/* ...and register our contact info with the HNP */
|
|
if (ORTE_SUCCESS != (rc = orte_rml_base_register_contact_info())) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
} else {
|
|
/* ndat != NULL means we are getting an update of RML info
|
|
* for the daemons - so update our contact info and routes
|
|
*/
|
|
orte_rml_base_contact_info_notify(ndat, NULL);
|
|
}
|
|
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
/* if ndat=NULL, then I can just ignore anything else - this is
|
|
* being called because there are places where other routing
|
|
* algos need to be called and we don't
|
|
*/
|
|
if (NULL == ndat) {
|
|
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
|
|
"%s routed_tree: no routing info provided for daemons",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
/* for any other job, extract the contact map from the launch
|
|
* message since it contains info from every daemon
|
|
*/
|
|
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
|
|
"%s routed_tree: extract proc routing info",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
|
|
|
|
values = (orte_gpr_value_t**)(ndat->values)->addr;
|
|
daemon.jobid = 0;
|
|
proc.jobid = job;
|
|
for (j=0, i=0; i < ndat->cnt && j < (ndat->values)->size; j++) { /* loop through all returned values */
|
|
if (NULL != values[j]) {
|
|
i++;
|
|
value = values[j];
|
|
|
|
if (NULL != value->tokens) {
|
|
/* this came from the globals container, so ignore it */
|
|
continue;
|
|
}
|
|
|
|
/* this must have come from one of the process containers, so it must
|
|
* contain data for a proc structure - extract what we need
|
|
*/
|
|
if (ORTE_SUCCESS != (rc = orte_odls.extract_proc_map_info(&daemon, &proc, value))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
|
|
if (0 != orte_ns.compare_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_NAME, &daemon)) {
|
|
/* Setup the route to the remote proc via its daemon */
|
|
if (ORTE_SUCCESS != (rc = orte_routed_tree_update_route(&proc, &daemon))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
} else {
|
|
/* setup the route for my own procs as they may not have talked
|
|
* to me yet - if they have, this will simply overwrite the existing
|
|
* entry, so no harm done
|
|
*/
|
|
if (ORTE_SUCCESS != (rc = orte_routed_tree_update_route(&proc, &proc))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
|
|
"%s routed_tree: completed init routes",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
|
|
/* I must be a proc - just setup my route to the local daemon */
|
|
{
|
|
int id;
|
|
char *rml_uri;
|
|
|
|
if (ORTE_EQUAL == orte_dss.compare(ORTE_NAME_INVALID, &orte_process_info.my_daemon, ORTE_NAME)) {
|
|
/* the daemon wasn't previously defined, so look for it */
|
|
id = mca_base_param_register_string("orte", "local_daemon", "uri", NULL, NULL);
|
|
mca_base_param_lookup_string(id, &rml_uri);
|
|
if (NULL == rml_uri) {
|
|
/* in this module, we absolutely MUST have this information - if
|
|
* we didn't get it, then error out
|
|
*/
|
|
opal_output(0, "%s ERROR: Failed to identify the local daemon's URI",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
|
|
opal_output(0, "%s ERROR: This is a fatal condition when the tree router",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
|
|
opal_output(0, "%s ERROR: has been selected - either select the unity router",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
|
|
opal_output(0, "%s ERROR: or ensure that the local daemon info is provided",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
|
|
return ORTE_ERR_FATAL;
|
|
}
|
|
/* Set the contact info in the RML - this won't actually establish
|
|
* the connection, but just tells the RML how to reach the daemon
|
|
* if/when we attempt to send to it
|
|
*/
|
|
if (ORTE_SUCCESS != (rc = orte_rml.set_contact_info(rml_uri))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
free(rml_uri);
|
|
return(rc);
|
|
}
|
|
/* extract the daemon's name so we can update the routing table */
|
|
if (ORTE_SUCCESS != (rc = orte_rml_base_parse_uris(rml_uri, &orte_process_info.my_daemon, NULL))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
free(rml_uri);
|
|
return rc;
|
|
}
|
|
free(rml_uri); /* done with this */
|
|
}
|
|
|
|
/* setup the route to all other procs to flow through the daemon */
|
|
if (ORTE_SUCCESS != (rc = orte_routed_tree_update_route(ORTE_NAME_WILDCARD, &orte_process_info.my_daemon))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
/* set my proc state - this will fire the corresponding trigger so
|
|
* everything else in this procedure can happen. Note that it involves a
|
|
* a communication, which means that a connection to the local daemon
|
|
* will be initiated. Thus, the local daemon will subsequently know
|
|
* my contact info
|
|
*/
|
|
if (ORTE_SUCCESS != (rc = orte_smr.set_proc_state(ORTE_PROC_MY_NAME, ORTE_PROC_STATE_AT_STG1, 0))) {
|
|
ORTE_ERROR_LOG(rc);
|
|
return rc;
|
|
}
|
|
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
}
|
|
|
|
int orte_routed_tree_warmup_routes(void)
|
|
{
|
|
orte_std_cntr_t i, j, simultaneous, world_size, istop;
|
|
orte_process_name_t next, prev;
|
|
struct iovec inmsg[1], outmsg[1];
|
|
int ret;
|
|
|
|
if (orte_process_info.seed) {
|
|
/* the HNP does not need to participate as it already has
|
|
* a warmed-up connection to every daemon, so just return
|
|
*/
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
|
|
"%s routed_tree: warming up daemon wireup for %ld procs",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.num_procs));
|
|
|
|
world_size = orte_process_info.num_procs;
|
|
istop = world_size/2;
|
|
simultaneous = 1;
|
|
if (world_size < simultaneous) {
|
|
simultaneous = world_size;
|
|
}
|
|
next.jobid = 0;
|
|
prev.jobid = 0;
|
|
inmsg[0].iov_base = outmsg[0].iov_base = NULL;
|
|
inmsg[0].iov_len = outmsg[0].iov_len = 0;
|
|
|
|
for (i = 1 ; i <= istop ; i += simultaneous) {
|
|
#if 0
|
|
if (simultaneous > (istop - i)) {
|
|
/* only fill in the rest */
|
|
simultaneous = istop - i;
|
|
}
|
|
#endif
|
|
/* the HNP does not need to participate as it already has
|
|
* a warmed-up connection to every daemon, so we exclude
|
|
* vpid=0 from both send and receive
|
|
*/
|
|
for (j = 0 ; j < simultaneous ; ++j) {
|
|
next.vpid = (ORTE_PROC_MY_NAME->vpid + (i + j )) % world_size;
|
|
if (next.vpid == 0) {
|
|
continue;
|
|
}
|
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
|
|
"%s routed_tree: daemon wireup sending to %s",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(&next)));
|
|
|
|
/* sends do not wait for a match */
|
|
ret = orte_rml.send(&next,
|
|
outmsg,
|
|
1,
|
|
ORTE_RML_TAG_WIREUP,
|
|
0);
|
|
if (ret < 0) return ret;
|
|
}
|
|
for (j = 0 ; j < simultaneous ; ++j) {
|
|
prev.vpid = (ORTE_PROC_MY_NAME->vpid - (i + j) + world_size) % world_size;
|
|
if (prev.vpid == 0) {
|
|
continue;
|
|
}
|
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
|
|
"%s routed_tree: daemon wireup recving from %s",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(&prev)));
|
|
|
|
ret = orte_rml.recv(&prev,
|
|
inmsg,
|
|
1,
|
|
ORTE_RML_TAG_WIREUP,
|
|
0);
|
|
if (ret < 0) return ret;
|
|
}
|
|
}
|
|
return ORTE_SUCCESS;
|
|
|
|
}
|