such a way that converter will not be able to pack some of it. This commit adds
handling of such cases. If converter can't pack any data for a BTL the data is
sent over another BTL that has data to send.
This commit was SVN r16493.
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 The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info.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The other stage gates have been replaced by orte_grpcomm barrier functions.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.
no more work associated with this request. No more outstanding completions or
packets and send scheduling isn't running in another thread.
This commit was SVN r16013.
the ompi_convertor_need_buffers function to only return 0 if the convertor
is homogeneous (which it never does on the trunk, but does to on v1.2, but
that's a different issue). Only enable the heterogeneous rdma code for
a btl if it supports it (via a flag), as some btls need some work for this
to work properly. Currently only TCP and OpenIB extensively tested
This commit was SVN r15990.
side too. Otherwise a content of the recvreq->req_rdma array is replaced later
without freeing previous content and refcount on registration in mpool become
wrong.
This commit was SVN r15978.
from the posted generic_recv-queue.
- Move the PERUSE_COMM_MSG_MATCH_POSTED_REQ from
MCA_PML_OB1_RECV_REQUEST_MATCHED to
mca_pml_ob1_recv_frag_match() as suggested by Terry Dontje
Only post, if this is not a probe/iprobe request.
- Do not post PERUSE_COMM_REQ_MATCH_UNEX for probes / iprobes and
do in correct order before PERUSE_COMM_MSG_REMOVE_FROM_UNEX_Q
This commit was SVN r15947.
PERUSE_COMM_REQ_ACTIVATE event.
Therefore move the PERUSE_TRACE_COMM_EVENT for this event from
MCA_PML_BASE_SEND_REQUEST_INIT / MCA_PML_BASE_RECV_REQUEST_INIT
to the proper places into pml_ob1_isend.c / pml_ob1_irecv.c right
after the MCA_PML_OB1_SEND_REQUEST_INIT /
MCA_PML_OB1_RECV_REQUEST_INIT.
This commit was SVN r15945.
on a parameter and are determined in runtime. r15346 removed calculation of
correct sizes for this structures. This patch adds it back and fixes trac:1116, #1114.
This commit was SVN r15932.
The following SVN revision numbers were found above:
r15346 --> open-mpi/ompi@433f8a7694
The following Trac tickets were found above:
Ticket 1116 --> https://svn.open-mpi.org/trac/ompi/ticket/1116
* 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
have to construct/destruct only once. Therefore, the construction will
happens before digging for a PML, while the destruction just before
finalizing the component.
Add some OPAL_LIKELY/OPAL_UNLIKELY.
This commit was SVN r15347.
receive queues are shared among all PMLs, they are declared in the base PML,
and the selected PML is in charge of initializing and releasing them.
The CM PML is slightly different compared with OB1 or DR. Internally it use
2 different types of requests: light and heavy. However, now with this patch
both types of requests are stored in the same queue, and cast appropriately
on the allocation macro. This means we might use less memory than we allocate,
but in exchange we got full support for most of the parallel debuggers.
Another thing with this patch, is that now for all PML (CM included) the basic
PML requests start with the same fields, and they are declared in the same order
in the request structure. Moreover, the fields have been moved in such a way
that only one volatile/atomic will exist per line of cache (hopefully).
This commit was SVN r15346.
than just the PML/BTLs these days. Also clean up the code so that it
handles the situation where not all nodes register information for a given
node (rather than just spinning until that node sends information, like
we do today).
Includes r15234 and r15265 from the /tmp/bwb-modex branch.
This commit was SVN r15310.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r15234
r15265
relative bandwidths of each BTL. Precalculate what part of a message should
be send via each BTL in advance instead of doing it during scheduling.
This commit was SVN r15248.
each BTL. Precalculate what part of a message should be send via each BTL in
advance instead of doing it during scheduling.
This commit was SVN r15247.
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.
