To use an MPI enabled application you must first load the appropriate module, the command:
module load openmpi/gcc
Will load the latest version of OpenMPI compiled using gcc and gfortran. All versions of OpenMPI on the cluster, starting with 1.8.4, are CUDA enabled. Loading OpenMPI will automatically load the appropriate version of CUDA. You would then run your program via the job scheduler like normal.
There are number of parameters that can be set for MPI runs to specify certain runtime behaviors of an application. These are discussed in more detail below. Some general notes are:
- OpenMPI will default to the fastest interface for a given job layout. If a job is set to run across multiple nodes, the Infiniband interface will automatically be set.
- For GPU runs the default byte-transport layer is smcuda. You should not override this (e.g. by setting --mca btl openib,tcp).
- You don't need to specify certain options when submitting MPI jobs to the scheduler (e.g. -np X, CUDA_VISIBLE_DEVICES).
- The job scheduler takes care of assigning hosts, so a hostfile is not needed (and is counter productive). See the documentation on SLURM.
Advanced Performance Considerations
Specifying Task and Process Layout
For some types of calculations how processes are distributed is critical to maximize performance. OpenMPI provides a number of options for mapping processes to resources. As an example, it may be ideal to pin processes to a specific CPU to improve communication with the GPUs. On Gemini, all GPUs are on the same PCIe bus as CPU socket 0. To force processes to be placed on those CPU cores the --map-by * and --bind-to * flags are passed to mpirun:
If you want to explicitly control which CPUs (or cores) are used and how to layout each rank you can use a rankfile. For example:
In the above example, the rank 0 would be put on node compute17, core 0 of CPU 1, rank 1 on compute17, core 1 of CPU 1, and so forth.
In general this level of specificity isn't needed and shouldn't be specified. One instance where this may be needed is discussed below.
Single-node GPU runs
Gemini has a total of 72 GPUs. Each compute node in the cluster has by default 4 GPUs assigned to it (with the possibility for up to 8). This means that in principle 72 independent GPU, 18 independent 4 GPU or 9 independent 8 GPU based jobs can be run simultaneously. Depending on the application or type of calculation any of those combinations may be beneficial, but is entirely dependent on the task.
For example, in Amber 14 most GPU calculations don't scale well past 4 GPUs. But if you have multiple independent jobs Amber runs those can be run simultaneously using either anywhere from 1-4 GPUs each.
Multi-node GPU runs
On Gemini each node has 4 GPUs attached to it. This can be increased on-the-fly by reconfiguring the attached PCIe device for up to 8 GPUs per node. When an application needs (and would benefit from even more GPUs in a single MPI run) the Mellanox OFED stack can be enabled on a subset of nodes and GPUdirect remote direct memory access (GRDMA) over Infiniband turned on. Under this usage case, GPUs on other nodes will communicate with each other over Infiniband via GRDMA to reduce latency between processes. Without GRDMA an MPI run will most probably not see any performance improvement when using GPUs on different nodes.
In this case, it is critical that processes be placed on the CPUs closest to the Infiniband interface. This is done specifically via a rankfile akin to the example above. In that case, ranks 0-7 will placed on one node, using CPU 1 and cores 0-7 on the CPU, ranks 8-15 on another node using CPU 1, cores 0-7 and so forth.
Manual Page for mpirun
Single Process Multiple Data (SPMD) Model:
mpirun [ options ] <program>
[ <args> ]
Multiple Instruction Multiple Data (MIMD) Model:
mpirun [ global_options
] [ local_options1 ]
<program1> [ <args1> ] : [ local_options2 ]
<program2> [ <args2> ] : ... :[ local_optionsN ]
<programN> [ <argsN> ]
Note that in both models, invoking mpirun via an absolute path name is equivalent to specifying the --prefix option with a <dir> value equivalent to the directory where mpirun resides, minus its last subdirectory.
% /usr/local/bin/mpirun ...
is equivalent to
% mpirun --prefix /usr/local
If you are simply looking for how to run an MPI application, you probably want to use a command line of the following form:
[ -np X ] [ --hostfile <filename> ] <program>
This will run X copies of <program> in your current run-time environment (if running under a supported resource manager, Open MPI’s mpirun will usually automatically use the corresponding resource manager process starter, as opposed to, for example, rsh or ssh, which require the use of a hostfile, or will default to running all X copies on the localhost), scheduling by default) in a round-robin fashion by CPU slot. See the rest of this page for more details.
Please note that mpirun automatically binds processes as of the start of the v1.8 series. Two binding patterns are used in the absence of any further directives:
- Bind to core:
- when the number of processes
is <= 2
- Bind to socket:
- when the number of processes is > 2
If your application
uses threads, then you probably want to ensure that you are either not bound at all (by specifying --bind-to none), or bound to multiple cores using an appropriate binding level or specific number of processing elements per application process.
mpirun will send the name of the directory where it was invoked on the local node to each of the remote nodes, and attempt to change to that directory. See the "Current Working Directory" section below for further details.
- The program executable. This
is identified as the first non-recognized argument to mpirun.
these run-time arguments to every new process. These must always be the
last arguments to mpirun. If an app context file is used, <args> will be ignored.
- -h, --help
- Display help for this command
- -q, --quiet
- Suppress informative
messages from orterun during application execution.
- -v, --verbose
- Be verbose
- -V, --version
- Print version number. If no other arguments are given, this
will also cause orterun to exit.
Use one of the following options to specify which hosts (nodes) of the cluster to run on. Note that as of the start of the v1.8 release, mpirun will launch a daemon onto each host in the allocation (as modified by the following options) at the very beginning of execution, regardless of whether or not application processes will eventually be mapped to execute there. This is done to allow collection of hardware topology information from the remote nodes, thus allowing us to map processes against known topology. However, it is a change from the behavior in prior releases where daemons were only launched after mapping was complete, and thus only occurred on nodes where application processes would actually be executing.
- -H, -host, --host <host1,host2,...,hostN>
- List of hosts on which
to invoke processes.
- -hostfile, --hostfile <hostfile>
- Provide a hostfile to
- -machinefile, --machinefile <machinefile>
- Synonym for -hostfile.
The following options specify the number of processes to launch. Note that none of the options imply a particular binding policy - e.g., requesting N processes for each socket does not imply that the processes will be bound to the socket.
- -c, -n, --n, -np <#>
- Run this many copies of the program on the given
nodes. This option indicates that the specified file is an executable program
and not an application context. If no value is provided for the number of
copies to execute (i.e., neither the "-np" nor its synonyms are provided on
the command line), Open MPI will automatically execute a copy of the program
on each process slot (see below for description of a "process slot"). This
feature, however, can only be used in the SPMD model and will return an
error (without beginning execution of the application) otherwise.-<>
N times the number of objects of the specified type on each node.
- On each node, launch this many processes times the
number of processor sockets on the node. The -npersocket option also turns
on the -bind-to-socket option. (deprecated in favor of --map-by ppr:n:socket)
- -npernode, --npernode <#pernode>
- On each node, launch this many processes.
(deprecated in favor of --map-by ppr:n:node)
- -pernode, --pernode
- On each node,
launch one process -- equivalent to -npernode 1. (deprecated in favor of --map-by
To map processes:
- --map-by <foo>
- Map to the specified object,
defaults to socket. Supported options include slot, hwthread, core, L1cache,
L2cache, L3cache, socket, numa, board, node, sequential, distance, and
ppr. Any object can include modifiers by adding a : and any combination
of PE=n (bind n processing elements to each proc), SPAN (load balance the
processes across the allocation), OVERSUBSCRIBE (allow more processes on
a node than processing elements), and NOOVERSUBSCRIBE. This includes PPR,
where the pattern would be terminated by another colon to separate it from
- -bycore, --bycore
- Map processes by core (deprecated in favor
of --map-by core)
- -bysocket, --bysocket
- Map processes by socket (deprecated
in favor of --map-by socket)
- -nolocal, --nolocal
- Do not run any copies of the
launched application on the same node as orterun is running. This option
will override listing the localhost with --host or any other host-specifying
- -nooversubscribe, --nooversubscribe
- Do not oversubscribe any nodes;
error (without starting any processes) if the requested number of processes
would cause oversubscription. This option implicitly sets "max_slots" equal
to the "slots" value for each node.
- -bynode, --bynode
- Launch processes one
per node, cycling by node in a round-robin fashion. This spreads processes
evenly among nodes and assigns MPI_COMM_WORLD ranks in a round-robin, "by
To order processes’ ranks in MPI_COMM_WORLD:
- --rank-by <foo>
- Rank in round-robin fashion according to the specified object, defaults
to slot. Supported options include slot, hwthread, core, L1cache, L2cache,
L3cache, socket, numa, board, and node.
For process binding:
- Bind processes to the specified object, defaults to core. Supported
options include slot, hwthread, core, l1cache, l2cache, l3cache, socket,
numa, board, and none.
- -cpus-per-proc, --cpus-per-proc <#perproc>
- Bind each process
to the specified number of cpus. (deprecated in favor of --map-by <obj>:PE=n)
- -cpus-per-rank, --cpus-per-rank <#perrank>
- Alias for -cpus-per-proc. (deprecated in
favor of --map-by <obj>:PE=n)
- -bind-to-core, --bind-to-core
- Bind processes to cores
(deprecated in favor of --bind-to core)
- -bind-to-socket, --bind-to-socket
- Bind processes
to processor sockets (deprecated in favor of --bind-to socket)
- Do not bind processes (deprecated in favor of --bind-to none)
- -report-bindings, --report-bindings
- Report any bindings for launched processes.
- -slot-list, --slot-list <slots>
- List of processor IDs to be used for binding
MPI processes. The specified bindings will be applied to all MPI processes.
See explanation below for syntax.
- -rf, --rankfile <rankfile>
- Provide a rankfile file.
To manage standard I/O:
- -output-filename, --output-filename
- Redirect the stdout, stderr, and stddiag of all processes to a
process-unique version of the specified filename. Any directories in the
filename will automatically be created. Each output file will consist of
filename.id, where the id will be the processes’ rank in MPI_COMM_WORLD,
left-filled with zero’s for correct ordering in listings.
- -stdin, --stdin <rank>
- The MPI_COMM_WORLD rank of the process that is to receive stdin. The default
is to forward stdin to MPI_COMM_WORLD rank 0, but this option can be used
to forward stdin to any process. It is also acceptable to specify none,
indicating that no processes are to receive stdin.
- -tag-output, --tag-output
- Tag each line of output to stdout, stderr, and stddiag with [jobid, MCW_rank]<stdxxx>
indicating the process jobid and MPI_COMM_WORLD rank of the process that
generated the output, and the channel which generated it.
- Timestamp each line of output to stdout, stderr, and stddiag.
- -xml, --xml
- Provide all output to stdout, stderr, and stddiag in an xml
- -xterm, --xterm <ranks>
- Display the output from the processes identified
by their MPI_COMM_WORLD ranks in separate xterm windows. The ranks are specified
as a comma-separated list of ranges, with a -1 indicating all. A separate
window will be created for each specified process. Note: xterm will normally
terminate the window upon termination of the process running within it.
However, by adding a "!" to the end of the list of specified ranks, the
proper options will be provided to ensure that xterm keeps the window open
after the process terminates, thus allowing you to see the process’ output.
Each xterm window will subsequently need to be manually closed. Note: In
some environments, xterm may require that the executable be in the user’s
path, or be specified in absolute or relative terms. Thus, it may be necessary
to specify a local executable as "./foo" instead of just "foo". If xterm
fails to find the executable, mpirun will hang, but still respond correctly
to a ctrl-c. If this happens, please check that the executable is being specified
correctly and try again.
To manage files and runtime environment:
- -path, --path <path>
- <path> that will be used when attempting to locate the requested
executables. This is used prior to using the local PATH setting.
- Prefix directory that will be used to set the PATH and LD_LIBRARY_PATH
on the remote node before invoking Open MPI or the target process. See
the "Remote Execution" section, below.
- Copy the specified
executable(s) to remote machines prior to starting remote processes. The
executables will be copied to the Open MPI session directory and will be
deleted upon completion of the job.
- --preload-files <files>
- Preload the comma
separated list of files to the current working directory of the remote
machines where processes will be launched prior to starting those processes.
- --preload-files-dest-dir <path>
- The destination directory to be used for preload-files,
if other than the current working directory. By default, the absolute and
relative paths provided by --preload-files are used.
- --tmpdir <dir>
- Set the root
for the session directory tree for mpirun only.
- -wd <dir>
- Synonym for -wdir.
- -wdir <dir>
- Change to the directory <dir> before the user’s program executes.
See the "Current Working Directory" section for notes on relative paths.
Note: If the -wdir option appears both on the command line and in an application
context, the context will take precedence over the command line. Thus, if
the path to the desired wdir is different on the backend nodes, then it
must be specified as an absolute path that is correct for the backend node.
- -x <env>
- Export the specified environment variables to the remote nodes
before executing the program. Only one environment variable can be specified
per -x option. Existing environment variables can be specified or new variable
names specified with corresponding values. For example: % mpirun -x
DISPLAY -x OFILE=/tmp/out ...
The parser for the -x option is not very sophisticated; it does not even understand quoted values. Users are advised to set variables in the environment, and then use -x to export (not define) them.
Setting MCA parameters:
- -gmca, --gmca <key> <value>
- Pass global MCA parameters that are applicable to
all contexts. <key> is the parameter name; <value> is the parameter value.
- -mca, --mca <key> <value>
- Send arguments to various MCA modules. See the "MCA"
- -debug, --debug
- Invoke the user-level debugger
indicated by the orte_base_user_debugger MCA parameter.
- -debugger, --debugger
- Sequence of debuggers to search for when --debug is used (i.e. a synonym for
orte_base_user_debugger MCA parameter).
- -tv, --tv
- Launch processes under
the TotalView debugger. Deprecated backwards compatibility flag. Synonym
There are also other options:
- Allow mpirun
to run when executed by the root user (mpirun defaults to aborting when
launched as the root user).
- -aborted, --aborted <#>
- Set the maximum number
of aborted processes to display.
- --app <appfile>
- Provide an appfile, ignoring
all other command line options.
- -cf, --cartofile <cartofile>
- Provide a cartography
- Indicates that multiple app_contexts are being provided that
are a mix of 32/64-bit binaries.
- -leave-session-attached, --leave-session-attached
- Do not detach OmpiRTE daemons used by this application. This allows error
messages from the daemons as well as the underlying environment (e.g., when
failing to launch a daemon) to be output.
- -ompi-server, --ompi-server <uri or
- Specify the URI of the Open MPI server (or the mpirun to be used as
the server) , the name of the file (specified as file:filename) that contains
that info, or the PID (specified as pid:#) of the mpirun to be used as
- Print out mpirun’s PID during startup. The channel must
be either a ’-’ to indi cate that the pid is to be output to stdout, a ’+’ to
indicate that the pid is to be outp ut to stderr, or a filename to which
the pid is to be written.
- -report-uri, --report-uri <channel>
- Print out mpirun’s
URI during startup. The channel must be either a ’-’ to indi cate that the
URI is to be output to stdout, a ’+’ to indicate that the URI is to be outp
ut to stderr, or a filename to which the URI is to be written.
- Pause mpirun before launching the job until ompi-server is
detected. This is useful in scripts where ompi-server may be started in the
background, followed immediately by an mpirun command that wishes to connect
to it. Mpirun will pause until either the specified ompi-server is contacted
or the server-wait-time is exceeded.
- -server-wait-time, --server-wait-time <secs>
- The max amount of time (in seconds) mpirun should wait for the ompi-server
to start. The default is 10 seconds.
The following options are useful for developers; they are not generally useful to most ORTE and/or MPI users:
- -d, --debug-devel
- Enable debugging of the OmpiRTE (the run-time layer in Open
MPI). This is not generally useful for most users.
debugging of any OmpiRTE daemons used by this application.
- Enable debugging of any OmpiRTE daemons used by this application, storing
output in files.
- -launch-agent, --launch-agent
- Name of the executable that
is to be used to start processes on the remote nodes. The default is "orted".
This option can be used to test new daemon concepts, or to pass options
back to the daemons without having mpirun itself see them. For example,
specifying a launch agent of orted -mca odls_base_verbose 5 allows the developer
to ask the orted for debugging output without clutter from mpirun itself.
- Disable the automatic --prefix behavior
There may be other options
listed with mpirun --help.
- The maximum number of seconds that mpirun (mpiexec) will run. After this many seconds, mpirun will abort the launched job and exit.
One invocation of mpirun starts an MPI application running under Open MPI. If the application is single process multiple data (SPMD), the application can be specified on the mpirun command line.
If the application is multiple instruction multiple data (MIMD), comprising of multiple programs, the set of programs and argument can be specified in one of two ways: Extended Command Line Arguments, and Application Context.
An application context describes the MIMD program set including all arguments in a separate file. This file essentially contains multiple mpirun command lines, less the command name
itself. The ability to specify different options for different instantiations of a program is another reason to use an application context.
Extended command line arguments allow for the description of the application layout on the command line using colons (:) to separate the specification of programs and arguments. Some options are globally set across all specified programs (e.g. --hostfile), while others are specific to a single program (e.g. -np).
Specifying Host Nodes
Host nodes can be identified on the mpirun command line with the -host option or in a hostfile.
- mpirun -H aa,aa,bb
- launches two processes on node aa and one on bb.
Or, consider the
% cat myhostfile
Here, we list both the host names (aa, bb, and cc) but also how many "slots" there are for each. Slots indicate how many processes can potentially execute on a node. For best performance, the number of slots may be chosen to be the number of cores on the node or the number of processor sockets. If the hostfile does not provide slots information, a default of 1 is assumed. When running under resource managers (e.g., SLURM, Torque, etc.), Open MPI will obtain both the hostnames and the number of slots directly from the resource manger.
- mpirun -hostfile myhostfile ./a.out
- will launch
two processes on each of the three nodes.
- mpirun -hostfile myhostfile -host
- will launch two processes, both on node aa.
- mpirun -hostfile myhostfile
-host dd ./a.out
- will find no hosts to run on and abort with an error. That
is, the specified host dd is not in the specified hostfile.
Specifying Number of Processes
As we have just seen, the number of processes to run can be set using the hostfile. Other mechanisms exist.
The number of processes launched can be specified as a multiple of the number of nodes or processor sockets available. For example,
- mpirun -H aa,bb -npersocket 2 ./a.out
processes 0-3 on node aa and process 4-7 on node bb, where aa and bb are
both dual-socket nodes. The -npersocket option also turns on the -bind-to-socket
option, which is discussed in a later section.
- mpirun -H aa,bb -npernode
- launches processes 0-1 on node aa and processes 2-3 on node bb.
-H aa,bb -npernode 1 ./a.out
- launches one process per host node.
- mpirun -H aa,bb
- is the same as -npernode 1.
Another alternative is to specify
the number of processes with the -np option. Consider now the hostfile
% cat myhostfile
- mpirun -hostfile myhostfile -np 6 ./a.out
- will launch processes 0-3
on node aa and processes 4-5 on node bb. The remaining slots in the hostfile
will not be used since the -np option indicated that only 6 processes should
Mapping Processes to Nodes: Using Policies
The examples above illustrate the default mapping of process processes to nodes. This mapping can also be controlled with various mpirun options that describe mapping policies. Consider the same hostfile as above, again with -np 6:
node aa node bb node cc
mpirun 0 1 2 3 4 5
mpirun --map-by node 0 3 1 4 2 5
mpirun -nolocal 0 1 2 3 4 5
The --map-by node option will load balance the processes across the available nodes, numbering each process in a round-robin fashion.
The -nolocal option prevents any processes from being mapped onto the local host (in this case node aa). While mpirun typically consumes few system resources, -nolocal can be helpful for launching very large jobs where mpirun may actually need to use noticeable amounts of memory and/or processing time.
Just as -np can specify fewer processes than there are slots, it can also oversubscribe the slots. For example, with the same hostfile:
- mpirun -hostfile myhostfile
-np 14 ./a.out
- will launch processes 0-3 on node aa, 4-7 on bb, and 8-11 on cc.
It will then add the remaining two processes to whichever nodes it chooses.
One can also specify limits to oversubscription. For example, with the same hostfile:
- mpirun -hostfile myhostfile -np 14 -nooversubscribe ./a.out
- will produce an error since -nooversubscribe prevents oversubscription.
Limits to oversubscription can also be specified in the hostfile itself:
% cat myhostfile
aa slots=4 max_slots=4
The max_slots field specifies such a limit. When it does, the slots value defaults to the limit. Now:
- mpirun -hostfile myhostfile -np 14 ./a.out
the first 12 processes to be launched as before, but the remaining two
processes will be forced onto node cc. The other two nodes are protected
by the hostfile against oversubscription by this job.
Using the --nooversubscribe option can be helpful since Open MPI currently does not get "max_slots" values from the resource manager.
Of course, -np can also be used with the -H or -host option. For example,
- mpirun -H aa,bb -np 8 ./a.out
- launches 8 processes.
Since only two hosts are specified, after the first two processes are
mapped, one to aa and one to bb, the remaining processes oversubscribe
the specified hosts.
And here is a MIMD example:
- mpirun -H aa -np 1 hostname
: -H bb,cc -np 2 uptime
- will launch process 0 running hostname on node aa
and processes 1 and 2 each running uptime on nodes bb and cc, respectively.
Mapping, Ranking, and Binding: Oh My!
Open MPI employs a three-phase procedure for assigning process locations and ranks:
- Assigns a default location
to each process
- Assigns an MPI_COMM_WORLD rank value to each process
- Constrains each process to run on specific processors
The mapping step is used to assign a default location to each process based on the mapper being employed. Mapping by slot, node, and sequentially results in the assignment of the processes to the node level. In contrast, mapping by object, allows the mapper to assign the process to an actual object on each node.
Note: the location assigned to the process is independent of where it will be bound - the assignment is used solely as input to the binding algorithm.
The mapping of process processes to nodes can be defined not just with general policies but also, if necessary, using arbitrary mappings that cannot be described by a simple policy. One can use the "sequential mapper," which reads the hostfile line by line, assigning processes to nodes in whatever order the hostfile specifies. Use the -mca rmaps seq option. For example, using the same hostfile as before:
mpirun -hostfile myhostfile
-mca rmaps seq ./a.out
will launch three processes, one on each of nodes aa, bb, and cc, respectively. The slot counts don’t matter; one process is launched per line on whatever node is listed on the line.
Another way to specify arbitrary mappings is with a rankfile, which gives you detailed control over process binding as well. Rankfiles are discussed below.
The second phase focuses on the ranking of the process within the job’s MPI_COMM_WORLD. Open MPI separates this from the mapping procedure to allow more flexibility in the relative placement of MPI processes. This is best illustrated by considering the following two cases where we used the —map-by ppr:2:socket option:
node aa node bb
rank-by core 0 1 ! 2 3 4 5 ! 6 7
rank-by socket 0 2 ! 1 3 4 6 ! 5 7
rank-by socket:span 0 4 ! 1 5 2 6 ! 3 7
Ranking by core and by slot provide the identical result - a simple progression of MPI_COMM_WORLD ranks across each node. Ranking by socket does a round-robin ranking within each node until all processes have been assigned an MCW rank, and then progresses to the next node. Adding the span modifier to the ranking directive causes the ranking algorithm to treat the entire allocation as a single entity - thus, the MCW ranks are assigned across all sockets before circling back around to the beginning.
The binding phase actually binds each process to a given set of processors. This can improve performance if the operating system is placing processes suboptimally. For example, it might oversubscribe some multi-core processor sockets, leaving other sockets idle; this can lead processes to contend unnecessarily for common resources. Or, it might spread processes out too widely; this can be suboptimal if application performance is sensitive to interprocess communication costs. Binding can also keep the operating system from migrating processes excessively, regardless of how optimally those processes were placed to begin with.
The processors to be used for binding can be identified in terms of topological groupings - e.g., binding to an l3cache will bind each process to all processors within the scope of a single L3 cache within their assigned location. Thus, if a process is assigned by the mapper to a certain socket, then a —bind-to l3cache directive will cause the process to be bound to the processors that share a single L3 cache within that socket.
To help balance loads, the binding directive uses a round-robin method when binding to levels lower than used in the mapper. For example, consider the case where a job is mapped to the socket level, and then bound to core. Each socket will have multiple cores, so if multiple processes are mapped to a given socket, the binding algorithm will assign each process located to a socket to a unique core in a round-robin manner.
Alternatively, processes mapped by l2cache and then bound to socket will simply be bound to all the processors in the socket where they are located. In this manner, users can exert detailed control over relative MCW rank location and binding.
Finally, --report-bindings can be used to report bindings.
As an example, consider a node with two processor sockets, each comprising four cores. We run mpirun with -np 4 --report-bindings and the following additional options:
% mpirun ... --map-by core --bind-to core[...] ... binding child [...,0] to cpus 0001 [...] ... binding child [...,1] to cpus 0002 [...] ... binding child [...,2] to cpus 0004 [...] ... binding child [...,3] to cpus 0008
% mpirun ... --map-by socket -0bind-to socket[...] ... binding child [...,0] to socket 0 cpus 000f [...] ... binding child [...,1] to socket 1 cpus 00f0 [...] ... binding child [...,2] to socket 0 cpus 000f [...] ... binding child [...,3] to socket 1 cpus 00f0
% mpirun ... --map-by core:PE=2 -bind-to core[...] ... binding child [...,0] to cpus 0003 [...] ... binding child [...,1] to cpus 000c [...] ... binding child [...,2] to cpus 0030 [...] ... binding child [...,3] to cpus 00c0
% mpirun ... --bind-to none
Here, --report-bindings shows the binding of each process as a mask. In the first case, the processes bind to successive cores as indicated by the masks 0001, 0002, 0004, and 0008. In the second case, processes bind to all cores on successive sockets as indicated by the masks 000f and 00f0. The processes cycle through the processor sockets in a round-robin fashion as many times as are needed. In the third case, the masks show us that 2 cores have been bound per process. In the fourth case, binding is turned off and no bindings are reported.
Open MPI’s support for process binding depends on the underlying operating system. Therefore, certain process binding options may not be available on every system.
Process binding can also be set with MCA parameters. Their usage is less convenient than that of mpirun options. On the other hand, MCA parameters can be set not only on the mpirun command line, but alternatively in a system or user mca-params.conf file or as environment variables, as described in the MCA section below. Some examples include:
mpirun option MCA parameter key value
--map-by core rmaps_base_mapping_policy core
--map-by socket rmaps_base_mapping_policy socket
--rank-by core rmaps_base_ranking_policy core
--bind-to core hwloc_base_binding_policy core
--bind-to socket hwloc_base_binding_policy socket
--bind-to none hwloc_base_binding_policy none
Rankfiles are text files that specify detailed information about how individual processes should be mapped to nodes, and to which processor(s) they should be bound. Each line of a rankfile specifies the location of one process (for MPI jobs, the process’ "rank" refers to its rank in MPI_COMM_WORLD). The general form of each line in the rankfile is:
rank <N>=<hostname> slot=<slot list>
$ cat myrankfile
rank 0=aa slot=1:0-2
rank 1=bb slot=0:0,1
rank 2=cc slot=1-2
$ mpirun -H aa,bb,cc,dd -rf myrankfile ./a.out
Rank 0 runs on node aa, bound to logical socket 1, cores
Rank 1 runs on node bb, bound to logical socket 0, cores 0 and 1.
Rank 2 runs on node cc, bound to logical cores 1 and 2.
Rankfiles can alternatively be used to specify physical processor locations. In this case, the syntax is somewhat different. Sockets are no longer recognized, and the slot number given must be the number of the physical PU as most OS’s do not assign a unique physical identifier to each core in the node. Thus, a proper physical rankfile looks something like the following:
$ cat myphysicalrankfile
rank 0=aa slot=1
rank 1=bb slot=8
rank 2=cc slot=6
This means that
Rank 0 will run on node aa, bound to the core that contains physical PU 1
Rank 1 will run on node bb, bound to the core that contains physical PU 8
Rank 2 will run on node cc, bound to the core that contains physical PU 6
Rankfiles are treated as logical by default, and the MCA parameter rmaps_rank_file_physical must be set to 1 to indicate that the rankfile is to be considered as physical.
The hostnames listed above are "absolute," meaning that actual resolveable hostnames are specified. However, hostnames can also be specified as "relative," meaning that they are specified in relation to an externally-specified list of hostnames (e.g., by mpirun’s --host argument, a hostfile, or a job scheduler).
The "relative" specification is of the form "+n<X>", where X is an integer specifying the Xth hostname in the set of all available hostnames, indexed from 0. For example:
$ cat myrankfile
rank 0=+n0 slot=1:0-2
rank 1=+n1 slot=0:0,1
rank 2=+n2 slot=1-2
$ mpirun -H aa,bb,cc,dd -rf myrankfile ./a.out
Starting with Open MPI v1.7, all socket/core slot locations are be specified as logical indexes (the Open MPI v1.6 series used physical indexes). You can use tools such as HWLOC’s "lstopo" to find the logical indexes of socket and cores.
If no relative or absolute path is specified for a file, Open MPI will first look for files by searching the directories specified by the --path option. If there is no --path option set or if the file is not found at the --path location, then Open MPI will search the user’s PATH environment variable as defined on the source node(s).
If a relative directory is specified, it must be relative to the initial working directory determined by the specific starter used. For example when using the rsh or ssh starters, the initial directory is $HOME by default. Other starters may set the initial directory to the current working directory from the invocation of mpirun.
Current Working Directory
The -wdir mpirun option (and its synonym, -wd) allows the user to change to an arbitrary directory before the program is invoked. It can also be used in application context files to specify working directories on specific nodes and/or for specific applications.
If the -wdir option appears both in a context file and on the command line, the context file directory will override the command line value.
If the -wdir option is specified, Open MPI will attempt to change to the specified directory on all of the remote nodes. If this fails, mpirun will abort.
If the -wdir option is not specified, Open MPI will send the directory name where mpirun was invoked to each of the remote nodes. The remote nodes will try to change to that directory. If they are unable (e.g., if the directory does not exist on that node), then Open MPI will use the default directory determined by the starter.
All directory changing occurs before the user’s program is invoked; it does not wait until MPI_INIT is called.
Open MPI directs UNIX standard input to /dev/null on all processes except the MPI_COMM_WORLD rank 0 process. The MPI_COMM_WORLD rank 0 process inherits standard input from mpirun. Note: The node that invoked mpirun need not be the same as the node where the MPI_COMM_WORLD rank 0 process resides. Open MPI handles the redirection of mpirun’s standard input to the rank 0 process.
Open MPI directs UNIX standard output and error from remote nodes to the node that invoked mpirun and prints it on the standard output/error of mpirun. Local processes inherit the standard output/error of mpirun and transfer to it directly. Thus it is possible to redirect standard I/O for Open MPI applications by using the typical shell redirection procedure on mpirun.
% mpirun -np 2 my_app < my_input > my_output
Note that in this example only the MPI_COMM_WORLD rank 0 process will receive the stream from my_input on stdin. The stdin on all the other nodes will be tied to /dev/null. However, the stdout from all nodes will be collected into the my_output file.