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 The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info.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The other stage gates have been replaced by orte_grpcomm barrier functions.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.
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.
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.
that allows to send any range of a request by send/recv instaed of RDMA
and use it to send data from the end of a request in pipeline protocol.
This commit was SVN r14841.
This is required to tighten up the BTL semantics. Ordering is not guaranteed,
but, if the BTL returns a order tag in a descriptor (other than
MCA_BTL_NO_ORDER) then we may request another descriptor that will obey
ordering w.r.t. to the other descriptor.
This will allow sane behavior for RDMA networks, where local completion of an
RDMA operation on the active side does not imply remote completion on the
passive side. If we send a FIN message after local completion and the FIN is
not ordered w.r.t. the RDMA operation then badness may occur as the passive
side may now try to deregister the memory and the RDMA operation may still be
pending on the passive side.
Note that this has no impact on networks that don't suffer from this
limitation as the ORDER tag can simply always be specified as
MCA_BTL_NO_ORDER.
This commit was SVN r14768.
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.
There is a binomial algorithm in the code (i.e., the HNP would send to a subset of the orteds, which then relay it on according to the typical log-2 algo), but that has a bug in it so the code won't let you select it even if you tried (and the mca param doesn't show, so you'd *really* have to try).
This also involved a slight change to the oob.xcast API, so propagated that as required.
Note: this has *only* been tested on rsh, SLURM, and Bproc environments (now that it has been transferred to the OMPI trunk, I'll need to re-test it [only done rsh so far]). It should work fine on any environment that uses the ORTE daemons - anywhere else, you are on your own... :-)
Also, correct a mistake where the orte_debug_flag was declared an int, but the mca param was set as a bool. Move the storage for that flag to the orte/runtime/params.c and orte/runtime/params.h files appropriately.
This commit was SVN r14475.
- Move the PML Modex stuff out of the BML -- Abstraction violation.
- Also fix the location of the add_procs with respect to the stage gates.
This commit was SVN r14422.
Fix for memory corruption in the restarted process stack. This stemed from
the brute force method we were previously using. This commit fixes this by
using a lighter weight solution focused in the r2 BML instead of above the PML.
This is a more efficient and flexible solution, and it solves the original
problem.
In the process I pulled out the ft_event function in the tcp BTL and r2 BML
into a set of *_ft.[c|h] files just to keep any updates to these code paths
as isolated as possible to make merging easier on everyone.
This commit was SVN r14371.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
The following Trac tickets were found above:
Ticket 977 --> https://svn.open-mpi.org/trac/ompi/ticket/977
Remove a redundant statement in the r2 BML.
This commit was SVN r14228.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
- Remove an old comment from crcp_base_fns.c
- Let ob1 have its very own ft_event function (which I'll fill in shortly)
- Make sure ob1 finalizes the bsend stuff so we don't leave a bunch of memory sitting around
- PML base - destruct the array upon finalize. Shrink the include search so it stops after finding a match
This commit was SVN r14222.
if less than or equal pml_ob1_unexpected_limit just buffer in the PML level recv
fragment else allocate a buffer via the bucket allocator
This commit was SVN r14117.
This merge adds Checkpoint/Restart support to Open MPI. The initial
frameworks and components support a LAM/MPI-like implementation.
This commit follows the risk assessment presented to the Open MPI core
development group on Feb. 22, 2007.
This commit closes trac:158
More details to follow.
This commit was SVN r14051.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r13912
The following Trac tickets were found above:
Ticket 158 --> https://svn.open-mpi.org/trac/ompi/ticket/158
adding new procs that the remote proc's pml is the same as our local pml.
Turns the hangs from mismatched PMLs into an abort, which is better,
I think.
This commit was SVN r13582.
* Make sure that the pval always writes to the correct portion of the
lval. This only matters on 32 bit big endian machines.
* On 32 bit machines when assigning to pval, the other 4 bytes of lval
weren't being written, which could lead to bogus data
We use macros so that there aren't casts all over the code and the pval
assignment can occur to the correct 4 bytes. Refs trac:587
This commit was SVN r12974.
The following Trac tickets were found above:
Ticket 587 --> https://svn.open-mpi.org/trac/ompi/ticket/587
r12714) for supporting compilers / architectures with different
padding rules.
This commit was SVN r12749.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r12491
r12714
hits the buffer on the other side. For this kind of BTLs we need to send
FIN through the same BTL, PUT was performed with so network will handle
ordering for us. If we will use another BTL, receiver can get FIN before
data will hit the buffer and complete request prematurely. We mark such
problematic BTLs with MCA_BTL_FLAGS_FAKE_RDMA flag (this kind of RDMA
is really fake, because the real one guaranties that sender will see the
completion only after receiver's NIC confirmed that all the data was
received).
This commit was SVN r12732.
parameter. For optimisation purpose only this BTL is used to send packet
through instead of trying to send packets through all BTLs. But actually the
code was wrong. It simply used provided bml_btl and it may represent different
endpoint from packet's destination. The fixed code checks if packet's
destination is reachable through the BTL, finds appropriate bml_btl and only
then tries to send it through correct bml_btl.
This commit was SVN r12319.
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.
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.
interconnects that provide matching logic in the library.
Currently includes support for MX and some support for
Portals
* Fix overuse of proc_pml pointer on the ompi_proc structuer,
splitting into proc_pml for pml data and proc_bml for
the BML endpoint data
* bug fixes in bsend init code, which wasn't being used by
the OB1 or DR PMLs...
This commit was SVN r10642.
flag, new flags to be included when convertor is initialized
- modified pml/btl module defs and added stub functions for diagnostic
output routines to dump state of queues / endpoints
- updates to data reliability pml
This commit was SVN r9329.
- 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.
