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 The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info.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The other stage gates have been replaced by orte_grpcomm barrier functions.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.
simultaneously, but is doing it incorrectly. If the function is running already
for one communicator and it is called from another thread for other communicator
with lower cid the check comm->c_contextid != ompi_comm_lowest_cid()
will fail and the function will be executed for two different communicators by
two threads simultaneously. There is nothing in the algorithm that prevent it
from been running simultaneously for different communicators as far as I can see,
but ompi_comm_unregister_cid() assumes that it is always called for a communicator
with the lowest cid and this is not always the case. This patch removes bogus
lowest cid check and fix ompi_comm_register_cid() to properly remove cid from
the list.
This commit was SVN r16088.
used at nce (up to one unique collective module per collective function).
Matches r15795:15921 of the tmp/bwb-coll-select branch
This commit was SVN r15924.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r15795
r15921
fix some of the multi-threading problems for the cid allocation. Two bugs
specifically:
- since we do not have a queue for incoming fragments of unknown cid, we need
to synchronize all processes before exiting the communicator creation. This
synchronization was/is located in comm_activate, which was however too late
for the multi-threaded case. Thus, for multi-threaded scenarios we are now
synchronizing 'before' we allow another thread to enter the cid-allocation
loop.
- for synchronization, we used for the sake of simplicity allreduce
operations. It turns out, that these operations interefered with the
allreductions in the cid-allocation routine, which lead to non-sense results
in the cid-allocation and potentially to endless loops.
Multi-threaded communicator creation seems to work now, is however still 'very
very' slow. I think, the busy wait of threads is killing the performance of
the active threads in the cid allocation. But this is another topic.
This commit was SVN r15910.
* General TCP cleanup for OPAL / ORTE
* Simplifying the OOB by moving much of the logic into the RML
* Allowing the OOB RML component to do routing of messages
* Adding a component framework for handling routing tables
* Moving the xcast functionality from the OOB base to its own framework
Includes merge from tmp/bwb-oob-rml-merge revisions:
r15506, r15507, r15508, r15510, r15511, r15512, r15513
This commit was SVN r15528.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r15506
r15507
r15508
r15510
r15511
r15512
r15513
Cleanup ALL instances of output involving the printing of orte_process_name_t structures using the ORTE_NAME_ARGS macro so that the number of fields and type of data match. Replace those values with a new macro/function pair ORTE_NAME_PRINT that outputs a string (using the new thread safe data capability) so that any future changes to the printing of those structures can be accomplished with a change to a single point.
Note that I could not possibly find outputs that directly print the orte_process_name_t fields, but only dealt with those that used ORTE_NAME_ARGS. Hence, you may still have a few outputs that bark during compilation. Also, I could only verify those that fall within environments I can compile on, so other environments may yield some minor warnings.
This commit was SVN r15517.
threads to create new communicators. There is nothing in the standard about threading
and communicaotr functions, but as they include collective communications I expect the
same rules have to be applied. As such, on an incorrect MPI program we deadlock (!).
This commit was SVN r15456.
VxWorks. Still some issues remaining, I'm sure.
Refs trac:1010
This commit was SVN r15320.
The following Trac tickets were found above:
Ticket 1010 --> https://svn.open-mpi.org/trac/ompi/ticket/1010
single threaded builds. In its default configuration, all this does
is ensure that there's at least a good chance of threads building
based on non-threaded development (since the variable names will be
checked). There is also code to make sure that a "mutex" is never
"double locked" when using the conditional macro mutex operations.
This is off by default because there are a number of places in both
ORTE and OMPI where this alarm spews mega bytes of errors on a
simple test. So we have some work to do on our path towards
thread support.
Also removed the macro versions of the non-conditional thread locks,
as the only places they were used, the author of the code intended
to use the conditional thread locks. So now you have upper-case
macros for conditional thread locks and lowercase functions for
non-conditional locks. Simple, right? :).
This commit was SVN r15011.
symbols in them and environ is defined only in the final application
(probably in crt1.o). Apple provides a function for getting at the
environment, so use that instead if it's available.
This commit was SVN r14857.
The primary change that underlies all this is in the OOB. Specifically, the problem in the code until now has been that the OOB attempts to resolve an address when we call the "send" to an unknown recipient. The OOB would then wait forever if that recipient never actually started (and hence, never reported back its OOB contact info). In the case of an orted that failed to start, we would correctly detect that the orted hadn't started, but then we would attempt to order all orteds (including the one that failed to start) to die. This would cause the OOB to "hang" the system.
Unfortunately, revising how the OOB resolves addresses introduced a number of additional problems. Specifically, and most troublesome, was the fact that comm_spawn involved the immediate transmission of the rendezvous point from parent-to-child after the child was spawned. The current code used the OOB address resolution as a "barrier" - basically, the parent would attempt to send the info to the child, and then "hold" there until the child's contact info had arrived (meaning the child had started) and the send could be completed.
Note that this also caused comm_spawn to "hang" the entire system if the child never started... The app-failed-to-start helped improve that behavior - this code provides additional relief.
With this change, the OOB will return an ADDRESSEE_UNKNOWN error if you attempt to send to a recipient whose contact info isn't already in the OOB's hash tables. To resolve comm_spawn issues, we also now force the cross-sharing of connection info between parent and child jobs during spawn.
Finally, to aid in setting triggers to the right values, we introduce the "arith" API for the GPR. This function allows you to atomically change the value in a registry location (either divide, multiply, add, or subtract) by the provided operand. It is equivalent to first fetching the value using a "get", then modifying it, and then putting the result back into the registry via a "put".
This commit was SVN r14711.
creating ompi_mpi_comm_null, since the destructor of ompi_mpi_comm_null will
decrease the reference counter of ompi_mpi_group_null twice according to the
last fix of Mohamad.
Added also a lengthy comment in ompi_comm_finalize about why we do
not decrease the reference counters for ompi_mpi_comm_null,
ompi_mpi_group_null etc. for the parent
communicator, although we do increase it in ompi_comm_init
This commit was SVN r14210.
* Do not add new procs to the global list during modex callback or
when sharing orte names during accept/connect. For modex, we
cache the modex info for later, in case that proc ever does get
added to the global proc list. For accept/connect orte name
exchange between the roots, we only need the orte name, so no
need to add a proc structure anyway. The procs will be added
to the global process list during the proc exchange later in
the wireup process
* Rename proc_get_namebuf and proc_get_proclist to proc_pack
and proc_unpack and extend them to include all information
needed to build that proc struct on a remote node (which
includes ORTE name, architecture, and hostname). Change
unpack to call pml_add_procs for the entire list of new
procs at once, rather than one at a time.
* Remove ompi_proc_find_and_add from the public proc
interface and make it a private function. This function
would add a half-created proc to the global proc list, so
making it harder to call is a good thing.
This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else. Currently, this is enough to implement MPI semantics. We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.
Refs trac:564
This commit was SVN r12798.
The following Trac tickets were found above:
Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
* don't load data into a buffer until we have the data, as
the data contains some header information needed to
properly load the data
This commit was SVN r12792.
the same time, remove some of the MPI-related options from OPAL:
- provide mechanism to change at runtime whether sched_yield() should
be called when the progress engine is idle
- provide mechanism for changing the rate at which the event engine
is called when there are "no" users of the event engine (ie, when
using MPI but not TCP)
- fix some function names in the progress engine to better match
their intended use (and remove MPI naming scheme)
- remove progress_mpi_enable / progress_mpi_disable because
we can now use the functions to set the sched_yield and
tick rate interfaces
- rename opal_progress_events() to opal_progress_set_event_flag()
because the first really isn't descriptive of what the function
does and I always got confused by it
This commit was SVN r12645.
Accordingly, there are new APIs to the name service to support the ability to get a job's parent, root, immediate children, and all its descendants. In addition, the terminate_job, terminate_orted, and signal_job APIs for the PLS have been modified to accept attributes that define the extent of their actions. For example, doing a "terminate_job" with an attribute of ORTE_NS_INCLUDE_DESCENDANTS will terminate the given jobid AND all jobs that descended from it.
I have tested this capability on a MacBook under rsh, Odin under SLURM, and LANL's Flash (bproc). It worked successfully on non-MPI jobs (both simple and including a spawn), and MPI jobs (again, both simple and with a spawn).
This commit was SVN r12597.
all platforms. The only exceptions (and I will not deal with them
anytime soon) are on Windows:
- the write functions which require the length to be an int when it's
a size_t on all UNIX variants.
- all iovec manipulation functions where the iov_len is again an int
when it's a size_t on most of the UNIXes.
As these only happens on Windows, so I think we're set for now :)
This commit was SVN r12215.
Modify the mapper to better bookmark its stopping place each time, and to pick up the next time from there. This needs to be validated on a multi-node system.
Fix a major memory corruption problem in the registry put/get functions that was doing multiple free's. Not sure how valgrind missed this one, though it only occurred in specific circumstances (such as comm_spawn).
This commit was SVN r12179.
This change does a couple of things:
1. Since the USE_PARENT_ALLOC attribute is a directive about regarding allocation of resources to a job, it more properly should be an attribute of the RAS. Change the name to reflect that and move the attribute define to the ras_types.h file.
2. Add the attributes list to the RMAPS map_job interface. This provides us with the desired flexibility to dynamically specify directives for mapping. The system will - in the absence of any attribute-based directive - default to the values provided in the MCA parameters (either from environment or command-line interface).
This commit was SVN r12164.
Fix the problem observed by multiple people that comm_spawned children were (once again) being mapped onto the same nodes as their parents. This was caused by going through the RAS a second time, thus overwriting the mapper's bookkeeping that told RMAPS where it had left off.
To solve this - and to continue moving forward on the ORTE development - we introduce the concept of attributes to control the behavior of the RM frameworks. I defined the attributes and a list of attributes as new ORTE data types to make it easier for people to pass them around (since they are now fundamental to the system, and therefore we will be packing and unpacking them frequently). Thus, all the functions to manipulate attributes can be implemented and debugged in one place.
I used those capabilities in two places:
1. Added an attribute list to the rmgr.spawn interface.
2. Added an attribute list to the ras.allocate interface. At the moment, the only attribute I modified the various RAS components to recognize is the USE_PARENT_ALLOCATION one (as defined in rmgr_types.h).
So the RAS components now know how to reuse an allocation. I have debugged this under rsh, but it now needs to be tested on a wider set of platforms.
This commit was SVN r12138.
something went wrong. A positiv number > 0 is however a correct value (in
contrary to orte_rml.recv_buffer, which really returns ORTE_SUCCESS or an
error code).
Note: this part of the code is correct on 1.1 and 1.2 branch, no need to move
this change patch to the release branches.
This commit was SVN r11897.
long ago) supposed to be used as a cache for accessing the PML procs. But in
all of the PMLs the PML proc contain only one field i.e. a pointer to the ompi_proc.
This pointer can be accessed using the c_remote_group easily. Therefore, there is no
meaning of keeping the PML procs around. Slim fast commit ...
This commit was SVN r11730.
Other changes:
1. Remove the old xcpu components as they are not functional.
2. Fix a "bug" in orterun whereby we called dump_aborted_procs even when we normally terminated. There is still some kind of bug in this procedure, however, as we appear to be calling the orterun job_state_callback function every time a process terminates (instead of only once when they have all terminated). I'll continue digging into that one.
This will require an autogen/configure, I'm afraid.
This commit was SVN r11228.
Clean up the remainder of the size_t references in the runtime itself. Convert to orte_std_cntr_t wherever it makes sense (only avoid those places where the actual memory size is referenced).
Remove the obsolete oob barrier function (we actually obsoleted it a long time ago - just never bothered to clean it up).
I have done my best to go through all the components and catch everything, even if I couldn't test compile them since I wasn't on that type of system. Still, I cannot guarantee that problems won't show up when you test this on specific systems. Usually, these will just show as "warning: comparison between signed and unsigned" notes which are easily fixed (just change a size_t to orte_std_cntr_t).
In some places, people didn't use size_t, but instead used some other variant (e.g., I found several places with uint32_t). I tried to catch all of them, but...
Once we get all the instances caught and fixed, this should once and for all resolve many of the heterogeneity problems.
This commit was SVN r11204.
1. comm_spawn processes by default will inherit the "--prefix" from their parent job. Thus, the "--prefix" provided on the command line will be propagated automatically to any children.
2. application programs can override the default by providing their own "ompi_prefix" in the MPI_Info parameter passed to comm_spawn
This commit was SVN r11143.
make it consistent with the indenting in the rest of the file
(otherwise it was quite difficult to understand -- saw this while I
was reviewing 11039).
This commit was SVN r11042.
object if provided.
The associated value is a comma-separated list of hosts -- which must be
in the initial allocation -- and is used to populate the application
context map.
This commit was SVN r11039.
comment in ompi_comm_invalid() in
source:/trunk/ompi/communicator/communicator.h.
Short version:
- ompi_comm_invalid() returns TRUE for MPI_COMM_NULL
- therefore MPI_COMM_C2F needs to explicitly check for MPI_COMM_NULL
(because it uses ompi_comm_invalid())
- make ~20 MPI functions only call ompi_comm_invalid() instead of
calling ompi_comm_invalid() *and* checking for MPI_COMM_NULL (~40 MPI
functions already only called ompi_comm_invalid() -- we should be
consistent)
- similar issue for ompi_win_invalid(), so I added a cross-referencing
comment in win.h and fixed MPI_WIN_SET_NAME to only call
ompi_win_invalid() (and not check for MPI_WIN_NULL)
This commit was SVN r9970.
- My original patch stands: MPI_FINALIZE directly invokes the
attribute callbacks on MPI_COMM_SELF
- We added some user-level checks to ensure that they don't call
MPI_FINALIZE twice (this isn't really required, but it will prevent
whacky segv's -- they'll at least get a nice error message)
- Removed the attribute callbacks on MPI_COMM_SELF from
ompi_mpi_comm_finalize (i.e., we just moved them from
ompi_mpi_comm_finalize to ompi_mpi_finalize -- we just moved this
process up earlier in the MPI_FINALIZE sequence of events)
- Because there were so many conversations about this, here's the
rationale:
- MPI-2:4.8 says that we have to MPI_COMM_FREE MPI_COMM_SELF so that
the attribute callbacks are invoked.
- After considerable discussion, we came to the conclusion that
FREE'ing COMM_SELF is not the issue -- calling the callbacks is
the issue.
- So it is sufficent for MPI_FINALIZE to directly invoke these
attribute callbacks
- The attribute callbacks are *not* invoked on other communicators
because said communicators are not MPI_COMM_FREE'ed
This commit was SVN r9628.
version 1.12. As in the 2.0 everything related to windows and files has been removed
I prefer to add the complete files, so I have a trace in the SN for later.
This commit was SVN r9373.
- move files out of toplevel include/ and etc/, moving it into the
sub-projects
- rather than including config headers with <project>/include,
have them as <project>
- require all headers to be included with a project prefix, with
the exception of the config headers ({opal,orte,ompi}_config.h
mpi.h, and mpif.h)
This commit was SVN r8985.
do for mpirun/orterun. This will allow -mca btl foo,self to work as
expected when doing MPI_COMM_SPAWN and friends.
This should be pushed to the v1.0 branch
This commit was SVN r8170.
comm_fill_rest there is no need for calling ompi_set_group_rank, since
we know already the rank of the process in the new comm. In case the
process was not part of the new communicator (rank = MPI_UNDEFINED)
calling this function caused a segfault on some platforms.
This commit was SVN r8060.
originally suggested by Ralf Wildenhues, to try to speed autogen, configure,
and make (and possibly even make install). Use automake's include directive
to drastically reduce the number of Makefile files (although the number of
Makefile.am files is the same - most are just included in a top-level
Makefile.am). Also use an Automake SUBDIRs feature to eliminate the
dynamic-mca tree, which was no longer really needed. This makes adding
a framework easier (since you don't have to remember the dynamic-mca
tree) and makes building faster (as make doesn't have to recurse through
the dynamic-mca tree)
This commit was SVN r7777.
discussed and cleared with Edgar.
Ensure that only processes who will be in the new communicator call
the coll selection function. It is pointless (and Bad in some cases)
for processes who are not in the new communicator to try to select a
coll module for the new communicator.
This commit was SVN r7573.
bits to expose the low bits being set. We were casting from a size_t to a void*
which is not good when working with big endian machines.
This fix makes MPI 2 dynamics work on PPC 64 (tested with a Linux OS).
This commit was SVN r7538.
The following SVN revision numbers were found above:
r7537 --> open-mpi/ompi@fd45714c03
AM_INIT_AUTOMAKE, instead of the deprecated version.
* Work around dumbness in modern AC_INIT that requires the version
number to be set at autoconf time (instead of at configure time, as
it was before). Set the version number, minus the subversion r number,
at autoconf time. Override the internal variables to include the r
number (if needed) at configure time. Basically, the right thing
should always happen. The only place it might not is the version
reported as part of configure --help will not have an r number.
* Since AM_INIT_AUTOMAKE taks a list of options, no need to specify
them in all the Makefile.am files.
* Addes support for subdir-objects, meaning that object files are put
in the directory containing source files, even if the Makefile.am is
in another directory. This should start making it feasible to
reduce the number of Makefile.am files we have in the tree, which
will greatly reduce the time to run autogen and configure.
This commit was SVN r7211.
* don't know what I was thinking, but can't use the MCA_PML_CALL macro on
the two data values, as they don't have things that the macro can
expand into
This commit was SVN r6868.
the values to the PML structure. This will allow PMLs that want to do
hardware matching at the cost of a smaller range of valid tags and cids.
Updated all the places that used the MPI_TAG_UB_VALUE constant to instead
look at the pml struct.
This commit was SVN r6778.
* rename ompi_basename to opal_basename
* rename ompi bitop functions to opal
* rename ompi_cmd_line to opal_cmd_line
* rename ompi_sizet2int to opal_sizet2int
* rename orte_daemon_init to opal_daemon_init
* rename ompi_few to opal_few
This commit was SVN r6330.
