1. taking advantage of the fact that we no longer create the launch message via a GPR trigger. In earlier times, we had the GPR create the launch message based on a subscription. In that mode of operation, we could not guarantee the order in which the data was stored in the message - hence, we had no choice but to parse the message in a loop that checked each value against a list of possible "keys" until the corresponding value was found.
Now, however, we construct the message "by hand", so we know precisely what data is in each location in the message. Thus, we no longer need to send the character string "keys" for each data value any more. This represents a rather large savings in the message size - to give you an example, we typically would use a 30-char "key" for a 2-byte data value. As you can see, the overhead can become very large.
2. sending node-specific data only once. Again, because we used to construct the message via subscriptions that were done on a per-proc basis, the data for each node (e.g., the daemon's name, whether or not the node was oversubscribed) would be included in the data for each proc. Thus, the node-specific data was repeated for every proc.
Now that we construct the message "by hand", there is no reason to do this any more. Instead, we can insert the data for a specific node only once, and then provide the per-proc data for that node. We therefore not only save all that extra data in the message, but we also only need to parse the per-node data once.
The savings become significant at scale. Here is a comparison between the revised trunk and the trunk prior to this commit (all data was taken on odin, using openib, 64 nodes, unity message routing, tested with application consisting of mpi_init/mpi_barrier/mpi_finalize, all execution times given in seconds, all launch message sizes in bytes):
Per-node scaling, taken at 1ppn:
#nodes original trunk revised trunk
time size time size
1 0.10 819 0.09 564
2 0.14 1070 0.14 677
3 0.15 1321 0.14 790
4 0.15 1572 0.15 903
8 0.17 2576 0.20 1355
16 0.25 4584 0.21 2259
32 0.28 8600 0.27 4067
64 0.50 16632 0.39 7683
Per-proc scaling, taken at 64 nodes
ppn original trunk revised trunk
time size time size
1 0.50 16669 0.40 7720
2 0.55 32733 0.54 11048
3 0.87 48797 0.81 14376
4 1.0 64861 0.85 17704
Condensing those numbers, it appears we gained:
per-node message size: 251 bytes/node -> 113 bytes/node
per-proc message size: 251 bytes/proc -> 52 bytes/proc
per-job message size: 568 bytes/job -> 399 bytes/job
(job-specific data such as jobid, override oversubscribe flag, total #procs in job, total slots allocated)
The fact that the two pre-commit trunk numbers are the same confirms the fact that each proc was containing the node data as well. It isn't quite the 10x message reduction I had hoped to get, but it is significant and gives much better scaling.
Note that the timing info was, as usual, pretty chaotic - the numbers cited here were typical across several runs taken after the initial one to avoid NFS file positioning influences.
Also note that this commit removes the orte_process_info.vpid_start field and the handful of places that passed that useless value. By definition, all jobs start at vpid=0, so all we were doing is passing "0" around. In fact, many places simply hardwired it to "0" anyway rather than deal with it.
This commit was SVN r16428.
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.
Short description: major changes include -
1. singletons now fork/exec a local daemon to manage their operations.
2. the orte daemon code now resides in libopen-rte
3. daemons no longer use the orte triggering system during startup. Instead, they directly call back to their parent pls component to report ready to operate. A base function to count the callbacks has been provided.
I have modified all the pls components except xcpu and poe (don't understand either well enough to do it). Full functionality has been verified for rsh, SLURM, and TM systems. Compile has been verified for xgrid and gridengine.
This commit was SVN r15390.
The default odls has been updated and works fine. The process odls has been updated, but I could not verify its operation. The bproc ODLS has not been updated yet. Ralph will look at it soon.
This commit was SVN r15257.
1. generalize orte_rml.xcast to become a general broadcast-like messaging system. Messages can now be sent to any tag on the daemons or processes. Note that any message sent via xcast will be delivered to ALL processes in the specified job - you don't get to pick and choose. At a later date, we will introduce an augmented capability that will use the daemons as relays, but will allow you to send to a specified array of process names.
2. extended orte_rml.xcast so it supports more scalable message routing methodologies. At the moment, we support three: (a) direct, which sends the message directly to all recipients; (b) linear, which sends the message to the local daemon on each node, which then relays it to its own local procs; and (b) binomial, which sends the message via a binomial algo across all the daemons, each of which then relays to its own local procs. The crossover points between the algos are adjustable via MCA param, or you can simply demand that a specific algo be used.
3. orteds no longer exhibit two types of behavior: bootproxy or VM. Orteds now always behave like they are part of a virtual machine - they simply launch a job if mpirun tells them to do so. This is another step towards creating an "orteboot" functionality, but also provided a clean system for supporting message relaying.
Note one major impact of this commit: multiple daemons on a node cannot be supported any longer! Only a single daemon/node is now allowed.
This commit is known to break support for the following environments: POE, Xgrid, Xcpu, Windows. It has been tested on rsh, SLURM, and Bproc. Modifications for TM support have been made but could not be verified due to machine problems at LANL. Modifications for SGE have been made but could not be verified. The developers for the non-verified environments will be separately notified along with suggestions on how to fix the problems.
This commit was SVN r15007.
To be precise, given this hypothetical launching pattern:
host1: vpids 0, 2, 4, 6
host2: vpids 1, 3, 5, 7
The local_rank for these procs would be:
host1: vpids 0->local_rank 0, v2->lr1, v4->lr2, v6->lr3
host2: vpids 1->local_rank 0, v3->lr1, v5->lr2, v7->lr3
and the number of local procs on each node would be four. If vpid=0 then does a comm_spawn of one process on host1, the values of the parent job would remain unchanged. The local_rank of the child process would be 0 and its num_local_procs would be 1 since it is in a separate jobid.
I have verified this functionality for the rsh case - need to verify that slurm and other cases also get the right values. Some consolidation of common code is probably going to occur in the SDS components to make this simpler and more maintainable in the future.
This commit was SVN r14706.
1. no -np provided - put one proc/node across all allocated nodes
2. -np N provided, N > #nodes - we print a pretty error message and exit
3. -np N provided, N <= #nodes - put one proc/node across N nodes
I also added a new orte constant (ORTE_ERR_SILENT) that allows us to pass up the chain that an error was encountered, but NOT print ORTE_ERROR_LOG messages. This is intended to be used for cases where the error we encounter is NOT an orte error, but rather is one associated with incorrect user input (e.g., the preceding case 2). In such cases, there is no point in printing an ORTE_ERROR_LOG chain of messages as it isn't an orte error.
This commit was SVN r12821.
Obviously, people like bproc will have to get the app_num via another avenue...but that's a problem for another day. Several options are easily available.
This commit was SVN r12788.
make this warning-proof, loop over the uint64_ts as an array of integers
and use %x. The final string is just as random and formatted exactly
the same, so we're all good in that department.
Refs trac:655
This commit was SVN r12742.
The following Trac tickets were found above:
Ticket 655 --> https://svn.open-mpi.org/trac/ompi/ticket/655
wider space than getpid()
* Include <time.h> to get time()'s prototype
* Fix typo that prevented using /dev/urandom on systems that had it
This commit was SVN r11780.
- use the OPAL functions for PATH and environment variables
- make all headers C++ friendly
- no unamed structures
- no implicit cast.
Plus a full implementation for the orte_wait functions.
This commit was SVN r11347.
different macros, one for each project. Therefore, now we have OPAL_DECLSPEC,
ORTE_DECLSPEC and OMPI_DECLSPEC. Please use them based on the sub-project.
This commit was SVN r11270.
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.
so that it does not return an error when no universe is passed to it.
Also put back in the 'Slots In Use' column as it is now working properly
per Ralphs recent ras commits. Still not sure what 'Slots Alloc' is meant
to represent, so left that as #if 0'd out for the moment.
This commit was SVN r10739.
The following SVN revision numbers were found above:
r10718 --> open-mpi/ompi@47eef2e002
Update the help text to report errors when not following that rule.
Also updated the RMAPS help text to reflect the reorganization of some of the round-robin code into the base.
The new functionality has been tested under Mac OS-X and on Odin using an MPI program. Both byslot and bynode mapping have been checked and verified. Operational support for other systems needs to be verified - I respectfully request people's help in doing so.
This commit was SVN r10708.
1. Modifies the RAS framework so it correctly stores and retrieves the actual slots in use, not just those that were allocated. Although the RAS node structure had storage for the number of slots in use, it turned out that the base function for storing and retrieving that information ignored what was in the field and simply set it equal to the number of slots allocated. This has now been fixed.
2. Modified the RMAPS framework so it updates the registry with the actual number of slots used by the mapping. Note that daemons are still NOT counted in this process as daemons are NOT mapped at this time. This will be fixed in 2.0, but will not be addressed in 1.x.
3. Added a new MCA parameter "rmaps_base_no_oversubscribe" that tells the system not to oversubscribe nodes even if the underlying environment permits it. The default is to oversubscribe if needed and the underlying environment permits it. I'm sure someone may argue "why would a user do that?", but it turns out that (looking ahead to dynamic resource reservations) sometimes users won't know how many nodes or slots they've been given in advance - this just allows them to say "hey, I'd rather not run if I didn't get enough".
4. Reorganizes the RMAPS framework to more easily support multiple components. A lot of the logic in the round_robin mapper was very valuable to any component - this has been moved to the base so others can take advantage of it.
5. Added a new test program "hello_nodename" - just does "hello_world" but also prints out the name of the node it is on.
6. Made the orte_ras_node_t object a full ORTE data type so it can more easily be copied, packed, etc. This proved helpful for the RMAPS code reorganization and might be of use elsewhere too.
This commit was SVN r10697.
After seeing the uglyness that is removing directories in the
codebase I decided to push down this to the OPAL by extending the
opal/os_create_dirpath.(c|h) to contain some more functionality.
In this process I renamed 'os_create_dirpath' to 'os_dirpath' since it
is a bit more general now.
Added a few functions to:
- check if an directory is empty
- check to see if the access permissions are set correctly
- destroy the directory at the end of the dirpath
- By using a caller callback function (a la Perl, I believe)
for every file, the caller can have fine grained control over
whether a specific file is deleted or not.
This simplifies things a bit for orte_session_dir_(finalize|cleanup)
as it should no longer contain any of this functionality, but uses
these functions to do the work.
From the external perspective nothing has changed, from the
developer point of view we have some cleaner, more generic code.
This commit was SVN r10640.
from the tmp/jjhursey-ft-cr branch.
In this commit we change the way universe names are created.
Before we by default first created "default-universe" then
if there was a conflict we created "default-universe-PID"
where PID is the PID of the HNP.
Now we create "default-universe-PID" all the time (when
a default universe name is used). This makes it much
easier when trying to find a HNP from an outside app
(e.g. orte-ps, orteconsole, ...)
This also adds a "search" function to find all of the
universes on the machine. This is useful in many contexts
when trying to find a persistent daemon or when trying to
connect to a HNP.
This commit also makes orte_universe_t an opal_object_t,
which is something that needed to happen, and only effected
the SDS in one of it's base functions.
I was asked to bring this over to aid in fixing orteconsole
and orteprobe. Due to the change of orte_universe_t to
an object orteprobe may need to be updated to reflect this
change. Since orteprobe needs to be looked at anyway I'll
leave this to Ralph to take care of.
*Note*:
These changes do not depend upon any of the FT work (but
the FT work does depend upon them). These were brought over
to help in fixing some of the ORTE tool set that require
the functionality layed out in this patch.
Testing:
Ran the 'ibm' tests before and after this change, and all was
as well as before the change. If anyone notices additional
irregularities in the system let me know. But none are expected.
This commit was SVN r10550.
- 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.
