Provided by: python-mpi4py-doc_3.0.3-4build2_all 
NAME
mpi4py - MPI for Python
Author Lisandro Dalcin
Contact
dalcinl@gmail.com
Web Site
https://bitbucket.org/mpi4py/mpi4py
Date February 28, 2020
Abstract
This document describes the MPI for Python package. MPI for Python provides bindings of
the Message Passing Interface (MPI) standard for the Python programming language, allowing
any Python program to exploit multiple processors.
This package is constructed on top of the MPI-1/2/3 specifications and provides an object
oriented interface which resembles the MPI-2 C++ bindings. It supports point-to-point
(sends, receives) and collective (broadcasts, scatters, gathers) communications of any
picklable Python object, as well as optimized communications of Python object exposing the
single-segment buffer interface (NumPy arrays, builtin bytes/string/array objects)
INTRODUCTION
Over the last years, high performance computing has become an affordable resource to many
more researchers in the scientific community than ever before. The conjunction of quality
open source software and commodity hardware strongly influenced the now widespread
popularity of Beowulf class clusters and cluster of workstations.
Among many parallel computational models, message-passing has proven to be an effective
one. This paradigm is specially suited for (but not limited to) distributed memory
architectures and is used in today’s most demanding scientific and engineering application
related to modeling, simulation, design, and signal processing. However, portable
message-passing parallel programming used to be a nightmare in the past because of the
many incompatible options developers were faced to. Fortunately, this situation
definitely changed after the MPI Forum released its standard specification.
High performance computing is traditionally associated with software development using
compiled languages. However, in typical applications programs, only a small part of the
code is time-critical enough to require the efficiency of compiled languages. The rest of
the code is generally related to memory management, error handling, input/output, and user
interaction, and those are usually the most error prone and time-consuming lines of code
to write and debug in the whole development process. Interpreted high-level languages can
be really advantageous for this kind of tasks.
For implementing general-purpose numerical computations, MATLAB [1] is the dominant
interpreted programming language. In the open source side, Octave and Scilab are well
known, freely distributed software packages providing compatibility with the MATLAB
language. In this work, we present MPI for Python, a new package enabling applications to
exploit multiple processors using standard MPI “look and feel” in Python scripts.
[1] MATLAB is a registered trademark of The MathWorks, Inc.
What is MPI?
MPI, [mpi-using] [mpi-ref] the Message Passing Interface, is a standardized and portable
message-passing system designed to function on a wide variety of parallel computers. The
standard defines the syntax and semantics of library routines and allows users to write
portable programs in the main scientific programming languages (Fortran, C, or C++).
Since its release, the MPI specification [mpi-std1] [mpi-std2] has become the leading
standard for message-passing libraries for parallel computers. Implementations are
available from vendors of high-performance computers and from well known open source
projects like MPICH [mpi-mpich] and Open MPI [mpi-openmpi].
What is Python?
Python is a modern, easy to learn, powerful programming language. It has efficient
high-level data structures and a simple but effective approach to object-oriented
programming with dynamic typing and dynamic binding. It supports modules and packages,
which encourages program modularity and code reuse. Python’s elegant syntax, together with
its interpreted nature, make it an ideal language for scripting and rapid application
development in many areas on most platforms.
The Python interpreter and the extensive standard library are available in source or
binary form without charge for all major platforms, and can be freely distributed. It is
easily extended with new functions and data types implemented in C or C++. Python is also
suitable as an extension language for customizable applications.
Python is an ideal candidate for writing the higher-level parts of large-scale scientific
applications [Hinsen97] and driving simulations in parallel architectures [Beazley97] like
clusters of PC’s or SMP’s. Python codes are quickly developed, easily maintained, and can
achieve a high degree of integration with other libraries written in compiled languages.
Related Projects
As this work started and evolved, some ideas were borrowed from well known MPI and Python
related open source projects from the Internet.
• OOMPI
• It has not relation with Python, but is an excellent object oriented approach to MPI.
• It is a C++ class library specification layered on top of the C bindings that
encapsulates MPI into a functional class hierarchy.
• It provides a flexible and intuitive interface by adding some abstractions, like Ports
and Messages, which enrich and simplify the syntax.
• Pypar
• Its interface is rather minimal. There is no support for communicators or process
topologies.
• It does not require the Python interpreter to be modified or recompiled, but does not
permit interactive parallel runs.
• General (picklable) Python objects of any type can be communicated. There is good
support for numeric arrays, practically full MPI bandwidth can be achieved.
• pyMPI
• It rebuilds the Python interpreter providing a built-in module for message passing. It
does permit interactive parallel runs, which are useful for learning and debugging.
• It provides an interface suitable for basic parallel programing. There is not full
support for defining new communicators or process topologies.
• General (picklable) Python objects can be messaged between processors. There is not
support for numeric arrays.
• Scientific Python
• It provides a collection of Python modules that are useful for scientific computing.
• There is an interface to MPI and BSP (Bulk Synchronous Parallel programming).
• The interface is simple but incomplete and does not resemble the MPI specification.
There is support for numeric arrays.
Additionally, we would like to mention some available tools for scientific computing and
software development with Python.
• NumPy is a package that provides array manipulation and computational capabilities
similar to those found in IDL, MATLAB, or Octave. Using NumPy, it is possible to write
many efficient numerical data processing applications directly in Python without using
any C, C++ or Fortran code.
• SciPy is an open source library of scientific tools for Python, gathering a variety of
high level science and engineering modules together as a single package. It includes
modules for graphics and plotting, optimization, integration, special functions, signal
and image processing, genetic algorithms, ODE solvers, and others.
• Cython is a language that makes writing C extensions for the Python language as easy as
Python itself. The Cython language is very close to the Python language, but Cython
additionally supports calling C functions and declaring C types on variables and class
attributes. This allows the compiler to generate very efficient C code from Cython code.
This makes Cython the ideal language for wrapping for external C libraries, and for fast
C modules that speed up the execution of Python code.
• SWIG is a software development tool that connects programs written in C and C++ with a
variety of high-level programming languages like Perl, Tcl/Tk, Ruby and Python. Issuing
header files to SWIG is the simplest approach to interfacing C/C++ libraries from a
Python module.
[mpi-std1]
MPI Forum. MPI: A Message Passing Interface Standard. International Journal of
Supercomputer Applications, volume 8, number 3-4, pages 159-416, 1994.
[mpi-std2]
MPI Forum. MPI: A Message Passing Interface Standard. High Performance Computing
Applications, volume 12, number 1-2, pages 1-299, 1998.
[mpi-using]
William Gropp, Ewing Lusk, and Anthony Skjellum. Using MPI: portable parallel
programming with the message-passing interface. MIT Press, 1994.
[mpi-ref]
Mark Snir, Steve Otto, Steven Huss-Lederman, David Walker, and Jack Dongarra. MPI -
The Complete Reference, volume 1, The MPI Core. MIT Press, 2nd. edition, 1998.
[mpi-mpich]
W. Gropp, E. Lusk, N. Doss, and A. Skjellum. A high-performance, portable
implementation of the MPI message passing interface standard. Parallel Computing,
22(6):789-828, September 1996.
[mpi-openmpi]
Edgar Gabriel, Graham E. Fagg, George Bosilca, Thara Angskun, Jack J. Dongarra,
Jeffrey M. Squyres, Vishal Sahay, Prabhanjan Kambadur, Brian Barrett, Andrew
Lumsdaine, Ralph H. Castain, David J. Daniel, Richard L. Graham, and Timothy S.
Woodall. Open MPI: Goals, Concept, and Design of a Next Generation MPI
Implementation. In Proceedings, 11th European PVM/MPI Users’ Group Meeting, Budapest,
Hungary, September 2004.
[Hinsen97]
Konrad Hinsen. The Molecular Modelling Toolkit: a case study of a large scientific
application in Python. In Proceedings of the 6th International Python Conference,
pages 29-35, San Jose, Ca., October 1997.
[Beazley97]
David M. Beazley and Peter S. Lomdahl. Feeding a large-scale physics application to
Python. In Proceedings of the 6th International Python Conference, pages 21-29, San
Jose, Ca., October 1997.
OVERVIEW
MPI for Python provides an object oriented approach to message passing which grounds on
the standard MPI-2 C++ bindings. The interface was designed with focus in translating MPI
syntax and semantics of standard MPI-2 bindings for C++ to Python. Any user of the
standard C/C++ MPI bindings should be able to use this module without need of learning a
new interface.
Communicating Python Objects and Array Data
The Python standard library supports different mechanisms for data persistence. Many of
them rely on disk storage, but pickling and marshaling can also work with memory buffers.
The pickle modules provide user-extensible facilities to serialize general Python objects
using ASCII or binary formats. The marshal module provides facilities to serialize
built-in Python objects using a binary format specific to Python, but independent of
machine architecture issues.
MPI for Python can communicate any built-in or user-defined Python object taking advantage
of the features provided by the pickle module. These facilities will be routinely used to
build binary representations of objects to communicate (at sending processes), and
restoring them back (at receiving processes).
Although simple and general, the serialization approach (i.e., pickling and unpickling)
previously discussed imposes important overheads in memory as well as processor usage,
especially in the scenario of objects with large memory footprints being communicated.
Pickling general Python objects, ranging from primitive or container built-in types to
user-defined classes, necessarily requires computer resources. Processing is also needed
for dispatching the appropriate serialization method (that depends on the type of the
object) and doing the actual packing. Additional memory is always needed, and if its total
amount is not known a priori, many reallocations can occur. Indeed, in the case of large
numeric arrays, this is certainly unacceptable and precludes communication of objects
occupying half or more of the available memory resources.
MPI for Python supports direct communication of any object exporting the single-segment
buffer interface. This interface is a standard Python mechanism provided by some types
(e.g., strings and numeric arrays), allowing access in the C side to a contiguous memory
buffer (i.e., address and length) containing the relevant data. This feature, in
conjunction with the capability of constructing user-defined MPI datatypes describing
complicated memory layouts, enables the implementation of many algorithms involving
multidimensional numeric arrays (e.g., image processing, fast Fourier transforms, finite
difference schemes on structured Cartesian grids) directly in Python, with negligible
overhead, and almost as fast as compiled Fortran, C, or C++ codes.
Communicators
In MPI for Python, MPI.Comm is the base class of communicators. The MPI.Intracomm and
MPI.Intercomm classes are sublcasses of the MPI.Comm class. The MPI.Comm.Is_inter()
method (and MPI.Comm.Is_intra(), provided for convenience but not part of the MPI
specification) is defined for communicator objects and can be used to determine the
particular communicator class.
The two predefined intracommunicator instances are available: MPI.COMM_SELF and
MPI.COMM_WORLD. From them, new communicators can be created as needed.
The number of processes in a communicator and the calling process rank can be respectively
obtained with methods MPI.Comm.Get_size() and MPI.Comm.Get_rank(). The associated process
group can be retrieved from a communicator by calling the MPI.Comm.Get_group() method,
which returns an instance of the MPI.Group class. Set operations with MPI.Group objects
like like MPI.Group.Union(), MPI.Group.Intersect() and MPI.Group.Difference() are fully
supported, as well as the creation of new communicators from these groups using
MPI.Comm.Create() and MPI.Comm.Create_group().
New communicator instances can be obtained with the MPI.Comm.Clone(), MPI.Comm.Dup() and
MPI.Comm.Split() methods, as well methods MPI.Intracomm.Create_intercomm() and
MPI.Intercomm.Merge().
Virtual topologies (MPI.Cartcomm, MPI.Graphcomm and MPI.Distgraphcomm classes, which are
specializations of the MPI.Intracomm class) are fully supported. New instances can be
obtained from intracommunicator instances with factory methods MPI.Intracomm.Create_cart()
and MPI.Intracomm.Create_graph().
Point-to-Point Communications
Point to point communication is a fundamental capability of message passing systems. This
mechanism enables the transmission of data between a pair of processes, one side sending,
the other receiving.
MPI provides a set of send and receive functions allowing the communication of typed data
with an associated tag. The type information enables the conversion of data
representation from one architecture to another in the case of heterogeneous computing
environments; additionally, it allows the representation of non-contiguous data layouts
and user-defined datatypes, thus avoiding the overhead of (otherwise unavoidable)
packing/unpacking operations. The tag information allows selectivity of messages at the
receiving end.
Blocking Communications
MPI provides basic send and receive functions that are blocking. These functions block
the caller until the data buffers involved in the communication can be safely reused by
the application program.
In MPI for Python, the MPI.Comm.Send(), MPI.Comm.Recv() and MPI.Comm.Sendrecv() methods of
communicator objects provide support for blocking point-to-point communications within
MPI.Intracomm and MPI.Intercomm instances. These methods can communicate memory buffers.
The variants MPI.Comm.send(), MPI.Comm.recv() and MPI.Comm.sendrecv() can communicate
general Python objects.
Nonblocking Communications
On many systems, performance can be significantly increased by overlapping communication
and computation. This is particularly true on systems where communication can be executed
autonomously by an intelligent, dedicated communication controller.
MPI provides nonblocking send and receive functions. They allow the possible overlap of
communication and computation. Non-blocking communication always come in two parts:
posting functions, which begin the requested operation; and test-for-completion functions,
which allow to discover whether the requested operation has completed.
In MPI for Python, the MPI.Comm.Isend() and MPI.Comm.Irecv() methods initiate send and
receive operations, respectively. These methods return a MPI.Request instance, uniquely
identifying the started operation. Its completion can be managed using the
MPI.Request.Test(), MPI.Request.Wait() and MPI.Request.Cancel() methods. The management of
MPI.Request objects and associated memory buffers involved in communication requires a
careful, rather low-level coordination. Users must ensure that objects exposing their
memory buffers are not accessed at the Python level while they are involved in nonblocking
message-passing operations.
Persistent Communications
Often a communication with the same argument list is repeatedly executed within an inner
loop. In such cases, communication can be further optimized by using persistent
communication, a particular case of nonblocking communication allowing the reduction of
the overhead between processes and communication controllers. Furthermore , this kind of
optimization can also alleviate the extra call overheads associated to interpreted,
dynamic languages like Python.
In MPI for Python, the MPI.Comm.Send_init() and MPI.Comm.Recv_init() methods create
persistent requests for a send and receive operation, respectively. These methods return
an instance of the MPI.Prequest class, a subclass of the MPI.Request class. The actual
communication can be effectively started using the MPI.Prequest.Start() method, and its
completion can be managed as previously described.
Collective Communications
Collective communications allow the transmittal of data between multiple processes of a
group simultaneously. The syntax and semantics of collective functions is consistent with
point-to-point communication. Collective functions communicate typed data, but messages
are not paired with an associated tag; selectivity of messages is implied in the calling
order. Additionally, collective functions come in blocking versions only.
The more commonly used collective communication operations are the following.
• Barrier synchronization across all group members.
• Global communication functions
• Broadcast data from one member to all members of a group.
• Gather data from all members to one member of a group.
• Scatter data from one member to all members of a group.
• Global reduction operations such as sum, maximum, minimum, etc.
In MPI for Python, the MPI.Comm.Bcast(), MPI.Comm.Scatter(), MPI.Comm.Gather(),
MPI.Comm.Allgather(), and MPI.Comm.Alltoall() MPI.Comm.Alltoallw() methods provide support
for collective communications of memory buffers. The lower-case variants MPI.Comm.bcast(),
MPI.Comm.scatter(), MPI.Comm.gather(), MPI.Comm.allgather() and MPI.Comm.alltoall() can
communicate general Python objects. The vector variants (which can communicate different
amounts of data to each process) MPI.Comm.Scatterv(), MPI.Comm.Gatherv(),
MPI.Comm.Allgatherv(), MPI.Comm.Alltoallv() and MPI.Comm.Alltoallw() are also supported,
they can only communicate objects exposing memory buffers.
Global reduction operations on memory buffers are accessible through the
MPI.Comm.Reduce(), MPI.Comm.Reduce_scatter, MPI.Comm.Allreduce(), MPI.Intracomm.Scan() and
MPI.Intracomm.Exscan() methods. The lower-case variants MPI.Comm.reduce(),
MPI.Comm.allreduce(), MPI.Intracomm.scan() and MPI.Intracomm.exscan() can communicate
general Python objects; however, the actual required reduction computations are performed
sequentially at some process. All the predefined (i.e., MPI.SUM, MPI.PROD, MPI.MAX, etc.)
reduction operations can be applied.
Dynamic Process Management
In the context of the MPI-1 specification, a parallel application is static; that is, no
processes can be added to or deleted from a running application after it has been started.
Fortunately, this limitation was addressed in MPI-2. The new specification added a process
management model providing a basic interface between an application and external resources
and process managers.
This MPI-2 extension can be really useful, especially for sequential applications built on
top of parallel modules, or parallel applications with a client/server model. The MPI-2
process model provides a mechanism to create new processes and establish communication
between them and the existing MPI application. It also provides mechanisms to establish
communication between two existing MPI applications, even when one did not start the
other.
In MPI for Python, new independent process groups can be created by calling the
MPI.Intracomm.Spawn() method within an intracommunicator. This call returns a new
intercommunicator (i.e., an MPI.Intercomm instance) at the parent process group. The child
process group can retrieve the matching intercommunicator by calling the
MPI.Comm.Get_parent() class method. At each side, the new intercommunicator can be used to
perform point to point and collective communications between the parent and child groups
of processes.
Alternatively, disjoint groups of processes can establish communication using a
client/server approach. Any server application must first call the MPI.Open_port()
function to open a port and the MPI.Publish_name() function to publish a provided service,
and next call the MPI.Intracomm.Accept() method. Any client applications can first find a
published service by calling the MPI.Lookup_name() function, which returns the port where
a server can be contacted; and next call the MPI.Intracomm.Connect() method. Both
MPI.Intracomm.Accept() and MPI.Intracomm.Connect() methods return an MPI.Intercomm
instance. When connection between client/server processes is no longer needed, all of them
must cooperatively call the MPI.Comm.Disconnect() method. Additionally, server
applications should release resources by calling the MPI.Unpublish_name() and
MPI.Close_port() functions.
One-Sided Communications
One-sided communications (also called Remote Memory Access, RMA) supplements the
traditional two-sided, send/receive based MPI communication model with a one-sided,
put/get based interface. One-sided communication that can take advantage of the
capabilities of highly specialized network hardware. Additionally, this extension lowers
latency and software overhead in applications written using a shared-memory-like paradigm.
The MPI specification revolves around the use of objects called windows; they intuitively
specify regions of a process’s memory that have been made available for remote read and
write operations. The published memory blocks can be accessed through three functions for
put (remote send), get (remote write), and accumulate (remote update or reduction) data
items. A much larger number of functions support different synchronization styles; the
semantics of these synchronization operations are fairly complex.
In MPI for Python, one-sided operations are available by using instances of the MPI.Win
class. New window objects are created by calling the MPI.Win.Create() method at all
processes within a communicator and specifying a memory buffer . When a window instance is
no longer needed, the MPI.Win.Free() method should be called.
The three one-sided MPI operations for remote write, read and reduction are available
through calling the methods MPI.Win.Put(), MPI.Win.Get(), and MPI.Win.Accumulate()
respectively within a Win instance. These methods need an integer rank identifying the
target process and an integer offset relative the base address of the remote memory block
being accessed.
The one-sided operations read, write, and reduction are implicitly nonblocking, and must
be synchronized by using two primary modes. Active target synchronization requires the
origin process to call the MPI.Win.Start() and MPI.Win.Complete() methods at the origin
process, and target process cooperates by calling the MPI.Win.Post() and MPI.Win.Wait()
methods. There is also a collective variant provided by the MPI.Win.Fence() method.
Passive target synchronization is more lenient, only the origin process calls the
MPI.Win.Lock() and MPI.Win.Unlock() methods. Locks are used to protect remote accesses to
the locked remote window and to protect local load/store accesses to a locked local
window.
Parallel Input/Output
The POSIX standard provides a model of a widely portable file system. However, the
optimization needed for parallel input/output cannot be achieved with this generic
interface. In order to ensure efficiency and scalability, the underlying parallel
input/output system must provide a high-level interface supporting partitioning of file
data among processes and a collective interface supporting complete transfers of global
data structures between process memories and files. Additionally, further efficiencies can
be gained via support for asynchronous input/output, strided accesses to data, and control
over physical file layout on storage devices. This scenario motivated the inclusion in the
MPI-2 standard of a custom interface in order to support more elaborated parallel
input/output operations.
The MPI specification for parallel input/output revolves around the use objects called
files. As defined by MPI, files are not just contiguous byte streams. Instead, they are
regarded as ordered collections of typed data items. MPI supports sequential or random
access to any integral set of these items. Furthermore, files are opened collectively by a
group of processes.
The common patterns for accessing a shared file (broadcast, scatter, gather, reduction) is
expressed by using user-defined datatypes. Compared to the communication patterns of
point-to-point and collective communications, this approach has the advantage of added
flexibility and expressiveness. Data access operations (read and write) are defined for
different kinds of positioning (using explicit offsets, individual file pointers, and
shared file pointers), coordination (non-collective and collective), and synchronism
(blocking, nonblocking, and split collective with begin/end phases).
In MPI for Python, all MPI input/output operations are performed through instances of the
MPI.File class. File handles are obtained by calling the MPI.File.Open() method at all
processes within a communicator and providing a file name and the intended access mode.
After use, they must be closed by calling the MPI.File.Close() method. Files even can be
deleted by calling method MPI.File.Delete().
After creation, files are typically associated with a per-process view. The view defines
the current set of data visible and accessible from an open file as an ordered set of
elementary datatypes. This data layout can be set and queried with the MPI.File.Set_view()
and MPI.File.Get_view() methods respectively.
Actual input/output operations are achieved by many methods combining read and write calls
with different behavior regarding positioning, coordination, and synchronism. Summing up,
MPI for Python provides the thirty (30) methods defined in MPI-2 for reading from or
writing to files using explicit offsets or file pointers (individual or shared), in
blocking or nonblocking and collective or noncollective versions.
Environmental Management
Initialization and Exit
Module functions MPI.Init() or MPI.Init_thread() and MPI.Finalize() provide MPI
initialization and finalization respectively. Module functions MPI.Is_initialized() and
MPI.Is_finalized() provide the respective tests for initialization and finalization.
NOTE:
MPI_Init() or MPI_Init_thread() is actually called when you import the MPI module from
the mpi4py package, but only if MPI is not already initialized. In such case, calling
MPI.Init() or MPI.Init_thread() from Python is expected to generate an MPI error, and
in turn an exception will be raised.
NOTE:
MPI_Finalize() is registered (by using Python C/API function Py_AtExit()) for being
automatically called when Python processes exit, but only if mpi4py actually
initialized MPI. Therefore, there is no need to call MPI.Finalize() from Python to
ensure MPI finalization.
Implementation Information
• The MPI version number can be retrieved from module function MPI.Get_version(). It
returns a two-integer tuple (version,subversion).
• The MPI.Get_processor_name() function can be used to access the processor name.
• The values of predefined attributes attached to the world communicator can be obtained
by calling the MPI.Comm.Get_attr() method within the MPI.COMM_WORLD instance.
Timers
MPI timer functionalities are available through the MPI.Wtime() and MPI.Wtick() functions.
Error Handling
In order facilitate handle sharing with other Python modules interfacing MPI-based
parallel libraries, the predefined MPI error handlers MPI.ERRORS_RETURN and
MPI.ERRORS_ARE_FATAL can be assigned to and retrieved from communicators, windows and
files using methods MPI.{Comm|Win|File}.Set_errhandler() and
MPI.{Comm|Win|File}.Get_errhandler().
When the predefined error handler MPI.ERRORS_RETURN is set, errors returned from MPI calls
within Python code will raise an instance of the exception class MPI.Exception, which is a
subclass of the standard Python exception RuntimeError.
NOTE:
After import, mpi4py overrides the default MPI rules governing inheritance of error
handlers. The MPI.ERRORS_RETURN error handler is set in the predefined MPI.COMM_SELF
and MPI.COMM_WORLD communicators, as well as any new MPI.Comm, MPI.Win, or MPI.File
instance created through mpi4py. If you ever pass such handles to C/C++/Fortran library
code, it is recommended to set the MPI.ERRORS_ARE_FATAL error handler on them to ensure
MPI errors do not pass silently.
WARNING:
Importing with from mpi4py.MPI import * will cause a name clashing with the standard
Python Exception base class.
TUTORIAL
WARNING:
Under construction. Contributions very welcome!
MPI for Python supports convenient, pickle-based communication of generic Python object as
well as fast, near C-speed, direct array data communication of buffer-provider objects
(e.g., NumPy arrays).
• Communication of generic Python objects
You have to use all-lowercase methods (of the Comm class), like send(), recv(), bcast().
An object to be sent is passed as a paramenter to the communication call, and the
received object is simply the return value.
The isend() and irecv() methods return Request instances; completion of these methods
can be managed using the test() and wait() methods of the Request class.
The recv() and irecv() methods may be passed a buffer object that can be repeatedly used
to receive messages avoiding internal memory allocation. This buffer must be
sufficiently large to accommodate the transmitted messages; hence, any buffer passed to
recv() or irecv() must be at least as long as the pickled data transmitted to the
receiver.
Collective calls like scatter(), gather(), allgather(), alltoall() expect a single value
or a sequence of Comm.size elements at the root or all process. They return a single
value, a list of Comm.size elements, or None.
• Communication of buffer-like objects
You have to use method names starting with an upper-case letter (of the Comm class),
like Send(), Recv(), Bcast(), Scatter(), Gather().
In general, buffer arguments to these calls must be explicitly specified by using a
2/3-list/tuple like [data, MPI.DOUBLE], or [data, count, MPI.DOUBLE] (the former one
uses the byte-size of data and the extent of the MPI datatype to define count).
For vector collectives communication operations like Scatterv() and Gatherv(), buffer
arguments are specified as [data, count, displ, datatype], where count and displ are
sequences of integral values.
Automatic MPI datatype discovery for NumPy arrays and PEP-3118 buffers is supported, but
limited to basic C types (all C/C99-native signed/unsigned integral types and
single/double precision real/complex floating types) and availability of matching
datatypes in the underlying MPI implementation. In this case, the buffer-provider object
can be passed directly as a buffer argument, the count and MPI datatype will be
inferred.
Running Python scripts with MPI
Most MPI programs can be run with the command mpiexec. In practice, running Python
programs looks like:
$ mpiexec -n 4 python script.py
to run the program with 4 processors.
Point-to-Point Communication
• Python objects (pickle under the hood):
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
if rank == 0:
data = {'a': 7, 'b': 3.14}
comm.send(data, dest=1, tag=11)
elif rank == 1:
data = comm.recv(source=0, tag=11)
• Python objects with non-blocking communication:
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
if rank == 0:
data = {'a': 7, 'b': 3.14}
req = comm.isend(data, dest=1, tag=11)
req.wait()
elif rank == 1:
req = comm.irecv(source=0, tag=11)
data = req.wait()
• NumPy arrays (the fast way!):
from mpi4py import MPI
import numpy
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
# passing MPI datatypes explicitly
if rank == 0:
data = numpy.arange(1000, dtype='i')
comm.Send([data, MPI.INT], dest=1, tag=77)
elif rank == 1:
data = numpy.empty(1000, dtype='i')
comm.Recv([data, MPI.INT], source=0, tag=77)
# automatic MPI datatype discovery
if rank == 0:
data = numpy.arange(100, dtype=numpy.float64)
comm.Send(data, dest=1, tag=13)
elif rank == 1:
data = numpy.empty(100, dtype=numpy.float64)
comm.Recv(data, source=0, tag=13)
Collective Communication
• Broadcasting a Python dictionary:
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
if rank == 0:
data = {'key1' : [7, 2.72, 2+3j],
'key2' : ( 'abc', 'xyz')}
else:
data = None
data = comm.bcast(data, root=0)
• Scattering Python objects:
from mpi4py import MPI
comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
if rank == 0:
data = [(i+1)**2 for i in range(size)]
else:
data = None
data = comm.scatter(data, root=0)
assert data == (rank+1)**2
• Gathering Python objects:
from mpi4py import MPI
comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
data = (rank+1)**2
data = comm.gather(data, root=0)
if rank == 0:
for i in range(size):
assert data[i] == (i+1)**2
else:
assert data is None
• Broadcasting a NumPy array:
from mpi4py import MPI
import numpy as np
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
if rank == 0:
data = np.arange(100, dtype='i')
else:
data = np.empty(100, dtype='i')
comm.Bcast(data, root=0)
for i in range(100):
assert data[i] == i
• Scattering NumPy arrays:
from mpi4py import MPI
import numpy as np
comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
sendbuf = None
if rank == 0:
sendbuf = np.empty([size, 100], dtype='i')
sendbuf.T[:,:] = range(size)
recvbuf = np.empty(100, dtype='i')
comm.Scatter(sendbuf, recvbuf, root=0)
assert np.allclose(recvbuf, rank)
• Gathering NumPy arrays:
from mpi4py import MPI
import numpy as np
comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
sendbuf = np.zeros(100, dtype='i') + rank
recvbuf = None
if rank == 0:
recvbuf = np.empty([size, 100], dtype='i')
comm.Gather(sendbuf, recvbuf, root=0)
if rank == 0:
for i in range(size):
assert np.allclose(recvbuf[i,:], i)
• Parallel matrix-vector product:
from mpi4py import MPI
import numpy
def matvec(comm, A, x):
m = A.shape[0] # local rows
p = comm.Get_size()
xg = numpy.zeros(m*p, dtype='d')
comm.Allgather([x, MPI.DOUBLE],
[xg, MPI.DOUBLE])
y = numpy.dot(A, xg)
return y
MPI-IO
• Collective I/O with NumPy arrays:
from mpi4py import MPI
import numpy as np
amode = MPI.MODE_WRONLY|MPI.MODE_CREATE
comm = MPI.COMM_WORLD
fh = MPI.File.Open(comm, "./datafile.contig", amode)
buffer = np.empty(10, dtype=np.int)
buffer[:] = comm.Get_rank()
offset = comm.Get_rank()*buffer.nbytes
fh.Write_at_all(offset, buffer)
fh.Close()
• Non-contiguous Collective I/O with NumPy arrays and datatypes:
from mpi4py import MPI
import numpy as np
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
amode = MPI.MODE_WRONLY|MPI.MODE_CREATE
fh = MPI.File.Open(comm, "./datafile.noncontig", amode)
item_count = 10
buffer = np.empty(item_count, dtype='i')
buffer[:] = rank
filetype = MPI.INT.Create_vector(item_count, 1, size)
filetype.Commit()
displacement = MPI.INT.Get_size()*rank
fh.Set_view(displacement, filetype=filetype)
fh.Write_all(buffer)
filetype.Free()
fh.Close()
Dynamic Process Management
• Compute Pi - Master (or parent, or client) side:
#!/usr/bin/env python
from mpi4py import MPI
import numpy
import sys
comm = MPI.COMM_SELF.Spawn(sys.executable,
args=['cpi.py'],
maxprocs=5)
N = numpy.array(100, 'i')
comm.Bcast([N, MPI.INT], root=MPI.ROOT)
PI = numpy.array(0.0, 'd')
comm.Reduce(None, [PI, MPI.DOUBLE],
op=MPI.SUM, root=MPI.ROOT)
print(PI)
comm.Disconnect()
• Compute Pi - Worker (or child, or server) side:
#!/usr/bin/env python
from mpi4py import MPI
import numpy
comm = MPI.Comm.Get_parent()
size = comm.Get_size()
rank = comm.Get_rank()
N = numpy.array(0, dtype='i')
comm.Bcast([N, MPI.INT], root=0)
h = 1.0 / N; s = 0.0
for i in range(rank, N, size):
x = h * (i + 0.5)
s += 4.0 / (1.0 + x**2)
PI = numpy.array(s * h, dtype='d')
comm.Reduce([PI, MPI.DOUBLE], None,
op=MPI.SUM, root=0)
comm.Disconnect()
Wrapping with SWIG
• C source:
/* file: helloworld.c */
void sayhello(MPI_Comm comm)
{
int size, rank;
MPI_Comm_size(comm, &size);
MPI_Comm_rank(comm, &rank);
printf("Hello, World! "
"I am process %d of %d.\n",
rank, size);
}
• SWIG interface file:
// file: helloworld.i
%module helloworld
%{
#include <mpi.h>
#include "helloworld.c"
}%
%include mpi4py/mpi4py.i
%mpi4py_typemap(Comm, MPI_Comm);
void sayhello(MPI_Comm comm);
• Try it in the Python prompt:
>>> from mpi4py import MPI
>>> import helloworld
>>> helloworld.sayhello(MPI.COMM_WORLD)
Hello, World! I am process 0 of 1.
Wrapping with F2Py
• Fortran 90 source:
! file: helloworld.f90
subroutine sayhello(comm)
use mpi
implicit none
integer :: comm, rank, size, ierr
call MPI_Comm_size(comm, size, ierr)
call MPI_Comm_rank(comm, rank, ierr)
print *, 'Hello, World! I am process ',rank,' of ',size,'.'
end subroutine sayhello
• Compiling example using f2py
$ f2py -c --f90exec=mpif90 helloworld.f90 -m helloworld
• Try it in the Python prompt:
>>> from mpi4py import MPI
>>> import helloworld
>>> fcomm = MPI.COMM_WORLD.py2f()
>>> helloworld.sayhello(fcomm)
Hello, World! I am process 0 of 1.
MPI4PY.FUTURES
New in version 3.0.0.
This package provides a high-level interface for asynchronously executing callables on a
pool of worker processes using MPI for inter-process communication.
concurrent.futures
The mpi4py.futures package is based on concurrent.futures from the Python standard
library. More precisely, mpi4py.futures provides the MPIPoolExecutor class as a concrete
implementation of the abstract class Executor. The submit() interface schedules a
callable to be executed asynchronously and returns a Future object representing the
execution of the callable. Future instances can be queried for the call result or
exception. Sets of Future instances can be passed to the wait() and as_completed()
functions.
NOTE:
The concurrent.futures package was introduced in Python 3.2. A backport targeting
Python 2.7 is available on PyPI. The mpi4py.futures package uses concurrent.futures if
available, either from the Python 3 standard library or the Python 2.7 backport if
installed. Otherwise, mpi4py.futures uses a bundled copy of core functionality
backported from Python 3.5 to work with Python 2.7.
SEE ALSO:
Module concurrent.futures
Documentation of the concurrent.futures standard module.
MPIPoolExecutor
The MPIPoolExecutor class uses a pool of MPI processes to execute calls asynchronously. By
performing computations in separate processes, it allows to side-step the Global
Interpreter Lock but also means that only picklable objects can be executed and returned.
The __main__ module must be importable by worker processes, thus MPIPoolExecutor instances
may not work in the interactive interpreter.
MPIPoolExecutor takes advantage of the dynamic process management features introduced in
the MPI-2 standard. In particular, the MPI.Intracomm.Spawn() method of MPI.COMM_SELF() is
used in the master (or parent) process to spawn new worker (or child) processes running a
Python interpreter. The master process uses a separate thread (one for each
MPIPoolExecutor instance) to communicate back and forth with the workers. The worker
processes serve the execution of tasks in the main (and only) thread until they are
signaled for completion.
NOTE:
The worker processes must import the main script in order to unpickle any callable
defined in the __main__ module and submitted from the master process. Furthermore, the
callables may need access to other global variables. At the worker
processes,:mod:mpi4py.futures executes the main script code (using the runpy module)
under the __worker__ namespace to define the __main__ module. The __main__ and
__worker__ modules are added to sys.modules (both at the master and worker processes)
to ensure proper pickling and unpickling.
WARNING:
During the initial import phase at the workers, the main script cannot create and use
new MPIPoolExecutor instances. Otherwise, each worker would attempt to spawn a new pool
of workers, leading to infinite recursion. mpi4py.futures detects such recursive
attempts to spawn new workers and aborts the MPI execution environment. As the main
script code is run under the __worker__ namespace, the easiest way to avoid spawn
recursion is using the idiom if __name__ == '__main__': ... in the main script.
class mpi4py.futures.MPIPoolExecutor(max_workers=None, **kwargs)
An Executor subclass that executes calls asynchronously using a pool of at most
max_workers processes. If max_workers is None or not given, its value is
determined from the MPI4PY_MAX_WORKERS environment variable if set, or the MPI
universe size if set, otherwise a single worker process is spawned. If max_workers
is lower than or equal to 0, then a ValueError will be raised.
Other parameters:
• python_exe: Path to the Python interpreter executable used to spawn worker
processes, otherwise sys.executable is used.
• python_args: list or iterable with additional command line flags to pass to the
Python executable. Command line flags determined from inspection of sys.flags,
sys.warnoptions and sys._xoptions in are passed unconditionally.
• mpi_info: dict or iterable yielding (key, value) pairs. These (key, value) pairs
are passed (through an MPI.Info object) to the MPI.Intracomm.Spawn() call used to
spawn worker processes. This mechanism allows telling the MPI runtime system
where and how to start the processes. Check the documentation of the backend MPI
implementation about the set of keys it interprets and the corresponding format
for values.
• globals: dict or iterable yielding (name, value) pairs to initialize the main
module namespace in worker processes.
• main: If set to False, do not import the __main__ module in worker processes.
Setting main to False prevents worker processes from accessing definitions in the
parent __main__ namespace.
• path: list or iterable with paths to append to sys.path in worker processes to
extend the module search path.
• wdir: Path to set the current working directory in worker processes using
os.chdir(). The initial working directory is set by the MPI implementation.
Quality MPI implementations should honor a wdir info key passed through mpi_info,
although such feature is not mandatory.
• env: dict or iterable yielding (name, value) pairs with environment variables to
update os.environ in worker processes. The initial environment is set by the MPI
implementation. MPI implementations may allow setting the initial environment
through mpi_info, however such feature is not required nor recommended by the MPI
standard.
submit(func, *args, **kwargs)
Schedule the callable, func, to be executed as func(*args, **kwargs) and
returns a Future object representing the execution of the callable.
executor = MPIPoolExecutor(max_workers=1)
future = executor.submit(pow, 321, 1234)
print(future.result())
map(func, *iterables, timeout=None, chunksize=1, **kwargs)
Equivalent to map(func, *iterables) except func is executed asynchronously
and several calls to func may be made concurrently, out-of-order, in
separate processes. The returned iterator raises a TimeoutError if
__next__() is called and the result isn’t available after timeout seconds
from the original call to map(). timeout can be an int or a float. If
timeout is not specified or None, there is no limit to the wait time. If a
call raises an exception, then that exception will be raised when its value
is retrieved from the iterator. This method chops iterables into a number of
chunks which it submits to the pool as separate tasks. The (approximate)
size of these chunks can be specified by setting chunksize to a positive
integer. For very long iterables, using a large value for chunksize can
significantly improve performance compared to the default size of one. By
default, the returned iterator yields results in-order, waiting for
successive tasks to complete . This behavior can be changed by passing the
keyword argument unordered as True, then the result iterator will yield a
result as soon as any of the tasks complete.
executor = MPIPoolExecutor(max_workers=3)
for result in executor.map(pow, [2]*32, range(32)):
print(result)
starmap(func, iterable, timeout=None, chunksize=1, **kwargs)
Equivalent to itertools.starmap(func, iterable). Used instead of map() when
argument parameters are already grouped in tuples from a single iterable
(the data has been “pre-zipped”). map(func, *iterable) is equivalent to
starmap(func, zip(*iterable)).
executor = MPIPoolExecutor(max_workers=3)
iterable = ((2, n) for n in range(32))
for result in executor.starmap(pow, iterable):
print(result)
shutdown(wait=True)
Signal the executor that it should free any resources that it is using when
the currently pending futures are done executing. Calls to submit() and
map() made after shutdown() will raise RuntimeError.
If wait is True then this method will not return until all the pending
futures are done executing and the resources associated with the executor
have been freed. If wait is False then this method will return immediately
and the resources associated with the executor will be freed when all
pending futures are done executing. Regardless of the value of wait, the
entire Python program will not exit until all pending futures are done
executing.
You can avoid having to call this method explicitly if you use the with
statement, which will shutdown the executor instance (waiting as if
shutdown() were called with wait set to True).
import time
with MPIPoolExecutor(max_workers=1) as executor:
future = executor.submit(time.sleep, 2)
assert future.done()
bootup(wait=True)
Signal the executor that it should allocate eagerly any required resources
(in particular, MPI worker processes). If wait is True, then bootup() will
not return until the executor resources are ready to process submissions.
Resources are automatically allocated in the first call to submit(), thus
calling bootup() explicitly is seldom needed.
NOTE:
As the master process uses a separate thread to perform MPI communication with the
workers, the backend MPI implementation should provide support for MPI.THREAD_MULTIPLE.
However, some popular MPI implementations do not support yet concurrent MPI calls from
multiple threads. Additionally, users may decide to initialize MPI with a lower level
of thread support. If the level of thread support in the backend MPI is less than
MPI.THREAD_MULTIPLE, mpi4py.futures will use a global lock to serialize MPI calls. If
the level of thread support is less than MPI.THREAD_SERIALIZED, mpi4py.futures will
emit a RuntimeWarning.
WARNING:
If the level of thread support in the backend MPI is less than MPI.THREAD_SERIALIZED
(i.e, it is either MPI.THREAD_SINGLE or MPI.THREAD_FUNNELED), in theory mpi4py.futures
cannot be used. Rather than raising an exception, mpi4py.futures emits a warning and
takes a “cross-fingers” attitude to continue execution in the hope that serializing MPI
calls with a global lock will actually work.
MPICommExecutor
Legacy MPI-1 implementations (as well as some vendor MPI-2 implementations) do not support
the dynamic process management features introduced in the MPI-2 standard. Additionally,
job schedulers and batch systems in supercomputing facilities may pose additional
complications to applications using the MPI_Comm_spawn() routine.
With these issues in mind, mpi4py.futures supports an additonal, more traditional,
SPMD-like usage pattern requiring MPI-1 calls only. Python applications are started the
usual way, e.g., using the mpiexec command. Python code should make a collective call to
the MPICommExecutor context manager to partition the set of MPI processes within a MPI
communicator in one master processes and many workers processes. The master process gets
access to an MPIPoolExecutor instance to submit tasks. Meanwhile, the worker process
follow a different execution path and team-up to execute the tasks submitted from the
master.
Besides alleviating the lack of dynamic process managment features in legacy MPI-1 or
partial MPI-2 implementations, the MPICommExecutor context manager may be useful in
classic MPI-based Python applications willing to take advantage of the simple, task-based,
master/worker approach available in the mpi4py.futures package.
class mpi4py.futures.MPICommExecutor(comm=None, root=0)
Context manager for MPIPoolExecutor. This context manager splits a MPI
(intra)communicator comm (defaults to MPI.COMM_WORLD if not provided or None) in
two disjoint sets: a single master process (with rank root in comm) and the
remaining worker processes. These sets are then connected through an
intercommunicator. The target of the with statement is assigned either an
MPIPoolExecutor instance (at the master) or None (at the workers).
from mpi4py import MPI
from mpi4py.futures import MPICommExecutor
with MPICommExecutor(MPI.COMM_WORLD, root=0) as executor:
if executor is not None:
future = executor.submit(abs, -42)
assert future.result() == 42
answer = set(executor.map(abs, [-42, 42]))
assert answer == {42}
WARNING:
If MPICommExecutor is passed a communicator of size one (e.g., MPI.COMM_SELF), then the
executor instace assigned to the target of the with statement will execute all
submitted tasks in a single worker thread, thus ensuring that task execution still
progress asynchronously. However, the GIL will prevent the main and worker threads from
running concurrently in multicore processors. Moreover, the thread context switching
may harm noticeably the performance of CPU-bound tasks. In case of I/O-bound tasks, the
GIL is not usually an issue, however, as a single worker thread is used, it progress
one task at a time. We advice against using MPICommExecutor with communicators of size
one and suggest refactoring your code to use instead a ThreadPoolExecutor.
Command line
Recalling the issues related to the lack of support for dynamic process managment features
in MPI implementations, mpi4py.futures supports an alternative usage pattern where Python
code (either from scripts, modules, or zip files) is run under command line control of the
mpi4py.futures package by passing -m mpi4py.futures to the python executable. The
mpi4py.futures invocation should be passed a pyfile path to a script (or a
zipfile/directory containing a __main__.py file). Additionally, mpi4py.futures accepts -m
mod to execute a module named mod, -c cmd to execute a command string cmd, or even - to
read commands from standard input (sys.stdin). Summarizing, mpi4py.futures can be invoked
in the following ways:
• $ mpiexec -n numprocs python -m mpi4py.futures pyfile [arg] ...
• $ mpiexec -n numprocs python -m mpi4py.futures -m mod [arg] ...
• $ mpiexec -n numprocs python -m mpi4py.futures -c cmd [arg] ...
• $ mpiexec -n numprocs python -m mpi4py.futures - [arg] ...
Before starting the main script execution, mpi4py.futures splits MPI.COMM_WORLD in one
master (the process with rank 0 in MPI.COMM_WORLD) and 16 workers and connect them through
an MPI intercommunicator. Afterwards, the master process proceeds with the execution of
the user script code, which eventually creates MPIPoolExecutor instances to submit tasks.
Meanwhile, the worker processes follow a different execution path to serve the master.
Upon successful termination of the main script at the master, the entire MPI execution
environment exists gracefully. In case of any unhandled exception in the main script, the
master process calls MPI.COMM_WORLD.Abort(1) to prevent deadlocks and force termination of
entire MPI execution environment.
WARNING:
Running scripts under command line control of mpi4py.futures is quite similar to
executing a single-process application that spawn additional workers as required.
However, there is a very important difference users should be aware of. All
MPIPoolExecutor instances created at the master will share the pool of workers. Tasks
submitted at the master from many different executors will be scheduled for execution
in random order as soon as a worker is idle. Any executor can easily starve all the
workers (e.g., by calling MPIPoolExecutor.map() with long iterables). If that ever
happens, submissions from other executors will not be serviced until free workers are
available.
SEE ALSO:
python:using-on-cmdline
Documentation on Python command line interface.
Examples
The following julia.py script computes the Julia set and dumps an image to disk in binary
PGM format. The code starts by importing MPIPoolExecutor from the mpi4py.futures package.
Next, some global constants and functions implement the computation of the Julia set. The
computations are protected with the standard if __name__ == '__main__':... idiom. The
image is computed by whole scanlines submitting all these tasks at once using the map
method. The result iterator yields scanlines in-order as the tasks complete. Finally, each
scanline is dumped to disk.
julia.py
from mpi4py.futures import MPIPoolExecutor
x0, x1, w = -2.0, +2.0, 640*2
y0, y1, h = -1.5, +1.5, 480*2
dx = (x1 - x0) / w
dy = (y1 - y0) / h
c = complex(0, 0.65)
def julia(x, y):
z = complex(x, y)
n = 255
while abs(z) < 3 and n > 1:
z = z**2 + c
n -= 1
return n
def julia_line(k):
line = bytearray(w)
y = y1 - k * dy
for j in range(w):
x = x0 + j * dx
line[j] = julia(x, y)
return line
if __name__ == '__main__':
with MPIPoolExecutor() as executor:
image = executor.map(julia_line, range(h))
with open('julia.pgm', 'wb') as f:
f.write(b'P5 %d %d %d\n' % (w, h, 255))
for line in image:
f.write(line)
The recommended way to execute the script is using the mpiexec command specifying one MPI
process and (optional but recommended) the desired MPI universe size [1].
$ mpiexec -n 1 -usize 17 python julia.py
The mpiexec command launches a single MPI process (the master) running the Python
interpreter and executing the main script. When required, mpi4py.futures spawns 16
additional MPI processes (the children) to dynamically allocate the pool of workers. The
master submits tasks to the children and waits for the results. The children receive
incoming tasks, execute them, and send back the results to the master.
Alternatively, users may decide to execute the script in a more traditional way, that is,
all the MPI process are started at once. The user script is run under command line control
of mpi4py.futures passing the -m flag to the python executable.
$ mpiexec -n 17 python -m mpi4py.futures julia.py
As explained previously, the 17 processes are partitioned in one master and 16 workers.
The master process executes the main script while the workers execute the tasks submitted
from the master.
[1] This mpiexec invocation example using the -usize flag (alternatively, setting the
MPIEXEC_UNIVERSE_SIZE environment variable) assumes the backend MPI implementation is
an MPICH derivative using the Hydra process manager. In the Open MPI implementation,
the MPI universe size can be specified by setting the OMPI_UNIVERSE_SIZE environment
variable to a positive integer. Check the documentation of your actual MPI
implementation and/or batch system for the ways to specify the desired MPI universe
size.
MPI4PY.RUN
New in version 3.0.0.
At import time, mpi4py initializes the MPI execution environment calling MPI_Init_thread()
and installs an exit hook to automatically call MPI_Finalize() just before the Python
process terminates. Additionally, mpi4py overrides the default MPI.ERRORS_ARE_FATAL error
handler in favor of MPI.ERRORS_RETURN, which allows translating MPI errors in Python
exceptions. These departures from standard MPI behavior may be controversial, but are
quite convenient within the highly dynamic Python programming environment. Third-party
code using mpi4py can just from mpi4py import MPI and perform MPI calls without the
tedious initialization/finalization handling. MPI errors, once translated automatically
to Python exceptions, can be dealt with the common try…except…finally clauses; unhandled
MPI exceptions will print a traceback which helps in locating problems in source code.
Unfortunately, the interplay of automatic MPI finalization and unhandled exceptions may
lead to deadlocks. In unattended runs, these deadlocks will drain the battery of your
laptop, or burn precious allocation hours in your supercomputing facility.
Consider the following snippet of Python code. Assume this code is stored in a standard
Python script file and run with mpiexec in two or more processes.
from mpi4py import MPI
assert MPI.COMM_WORLD.Get_size() > 1
rank = MPI.COMM_WORLD.Get_rank()
if rank == 0:
1/0
MPI.COMM_WORLD.send(None, dest=1, tag=42)
elif rank == 1:
MPI.COMM_WORLD.recv(source=0, tag=42)
Process 0 raises ZeroDivisionError exception before performing a send call to process 1.
As the exception is not handled, the Python interpreter running in process 0 will proceed
to exit with non-zero status. However, as mpi4py installed a finalizer hook to call
MPI_Finalize() before exit, process 0 will block waiting for other processes to also enter
the MPI_Finalize() call. Meanwhile, process 1 will block waiting for a message to arrive
from process 0, thus never reaching to MPI_Finalize(). The whole MPI execution environment
is irremediably in a deadlock state.
To alleviate this issue, mpi4py offers a simple, alternative command line execution
mechanism based on using the -m flag and implemented with the runpy module. To use this
features, Python code should be run passing -m mpi4py in the command line invoking the
Python interpreter. In case of unhandled exceptions, the finalizer hook will call
MPI_Abort() on the MPI_COMM_WORLD communicator, thus effectively aborting the MPI
execution environment.
WARNING:
When a process is forced to abort, resources (e.g. open files) are not cleaned-up and
any registered finalizers (either with the atexit module, the Python C/API function
Py_AtExit(), or even the C standard library function atexit()) will not be executed.
Thus, aborting execution is an extremely impolite way of ensuring process termination.
However, MPI provides no other mechanism to recover from a deadlock state.
Interface options
The use of -m mpi4py to execute Python code on the command line resembles that of the
Python interpreter.
• mpiexec -n numprocs python -m mpi4py pyfile [arg] ...
• mpiexec -n numprocs python -m mpi4py -m mod [arg] ...
• mpiexec -n numprocs python -m mpi4py -c cmd [arg] ...
• mpiexec -n numprocs python -m mpi4py - [arg] ...
<pyfile>
Execute the Python code contained in pyfile, which must be a filesystem path
referring to either a Python file, a directory containing a __main__.py file, or a
zipfile containing a __main__.py file.
-m <mod>
Search sys.path for the named module mod and execute its contents.
-c <cmd>
Execute the Python code in the cmd string command.
- Read commands from standard input (sys.stdin).
SEE ALSO:
python:using-on-cmdline
Documentation on Python command line interface.
CITATION
If MPI for Python been significant to a project that leads to an academic publication,
please acknowledge that fact by citing the project.
• L. Dalcin, P. Kler, R. Paz, and A. Cosimo, Parallel Distributed Computing using Python,
Advances in Water Resources, 34(9):1124-1139, 2011.
http://dx.doi.org/10.1016/j.advwatres.2011.04.013
• L. Dalcin, R. Paz, M. Storti, and J. D’Elia, MPI for Python: performance improvements
and MPI-2 extensions, Journal of Parallel and Distributed Computing, 68(5):655-662,
2008. http://dx.doi.org/10.1016/j.jpdc.2007.09.005
• L. Dalcin, R. Paz, and M. Storti, MPI for Python, Journal of Parallel and Distributed
Computing, 65(9):1108-1115, 2005. http://dx.doi.org/10.1016/j.jpdc.2005.03.010
INSTALLATION
Requirements
You need to have the following software properly installed in order to build MPI for
Python:
• A working MPI implementation, preferably supporting MPI-3 and built with shared/dynamic
libraries.
NOTE:
If you want to build some MPI implementation from sources, check the instructions at
building-mpi in the appendix.
• Python 2.7, 3.3 or above.
NOTE:
Some MPI-1 implementations do require the actual command line arguments to be passed
in MPI_Init(). In this case, you will need to use a rebuilt, MPI-enabled, Python
interpreter executable. MPI for Python has some support for alleviating you from this
task. Check the instructions at python-mpi in the appendix.
Using pip or easy_install
If you already have a working MPI (either if you installed it from sources or by using a
pre-built package from your favourite GNU/Linux distribution) and the mpicc compiler
wrapper is on your search path, you can use pip:
$ [sudo] pip install mpi4py
or alternatively setuptools easy_install (deprecated):
$ [sudo] easy_install mpi4py
NOTE:
If the mpicc compiler wrapper is not on your search path (or if it has a different
name) you can use env to pass the environment variable MPICC providing the full path to
the MPI compiler wrapper executable:
$ [sudo] env MPICC=/path/to/mpicc pip install mpi4py
$ [sudo] env MPICC=/path/to/mpicc easy_install mpi4py
Using distutils
The MPI for Python package is available for download at the project website generously
hosted by Bitbucket. You can use curl or wget to get a release tarball.
• Using curl:
$ curl -O https://bitbucket.org/mpi4py/mpi4py/downloads/mpi4py-X.Y.tar.gz
• Using wget:
$ wget https://bitbucket.org/mpi4py/mpi4py/downloads/mpi4py-X.Y.tar.gz
After unpacking the release tarball:
$ tar -zxf mpi4py-X.Y.tar.gz
$ cd mpi4py-X.Y
the package is ready for building.
MPI for Python uses a standard distutils-based build system. However, some distutils
commands (like build) have additional options:
--mpicc=
Lets you specify a special location or name for the mpicc compiler wrapper.
--mpi= Lets you pass a section with MPI configuration within a special configuration file.
--configure
Runs exhaustive tests for checking about missing MPI types, constants, and
functions. This option should be passed in order to build MPI for Python against
old MPI-1 or MPI-2 implementations, possibly providing a subset of MPI-3.
If you use a MPI implementation providing a mpicc compiler wrapper (e.g., MPICH, Open
MPI), it will be used for compilation and linking. This is the preferred and easiest way
of building MPI for Python.
If mpicc is located somewhere in your search path, simply run the build command:
$ python setup.py build
If mpicc is not in your search path or the compiler wrapper has a different name, you can
run the build command specifying its location:
$ python setup.py build --mpicc=/where/you/have/mpicc
Alternatively, you can provide all the relevant information about your MPI implementation
by editing the file called mpi.cfg. You can use the default section [mpi] or add a new,
custom section, for example [other_mpi] (see the examples provided in the mpi.cfg file as
a starting point to write your own section):
[mpi]
include_dirs = /usr/local/mpi/include
libraries = mpi
library_dirs = /usr/local/mpi/lib
runtime_library_dirs = /usr/local/mpi/lib
[other_mpi]
include_dirs = /opt/mpi/include ...
libraries = mpi ...
library_dirs = /opt/mpi/lib ...
runtime_library_dirs = /op/mpi/lib ...
...
and then run the build command, perhaps specifying you custom configuration section:
$ python setup.py build --mpi=other_mpi
After building, the package is ready for install.
If you have root privileges (either by log-in as the root user of by using sudo) and you
want to install MPI for Python in your system for all users, just do:
$ python setup.py install
The previous steps will install the mpi4py package at standard location
prefix/lib/pythonX.X/site-packages.
If you do not have root privileges or you want to install MPI for Python for your private
use, just do:
$ python setup.py install --user
Testing
To quickly test the installation:
$ mpiexec -n 5 python -m mpi4py.bench helloworld
Hello, World! I am process 0 of 5 on localhost.
Hello, World! I am process 1 of 5 on localhost.
Hello, World! I am process 2 of 5 on localhost.
Hello, World! I am process 3 of 5 on localhost.
Hello, World! I am process 4 of 5 on localhost.
If you installed from source, issuing at the command line:
$ mpiexec -n 5 python demo/helloworld.py
or (in the case of ancient MPI-1 implementations):
$ mpirun -np 5 python `pwd`/demo/helloworld.py
will launch a five-process run of the Python interpreter and run the test script
demo/helloworld.py from the source distribution.
You can also run all the unittest scripts:
$ mpiexec -n 5 python test/runtests.py
or, if you have nose unit testing framework installed:
$ mpiexec -n 5 nosetests -w test
or, if you have py.test unit testing framework installed:
$ mpiexec -n 5 py.test test/
APPENDIX
MPI-enabled Python interpreter
WARNING:
These days it is no longer required to use the MPI-enabled Python interpreter in
most cases, and, therefore, is not built by default anymore because it is too
difficult to reliably build a Python interpreter across different distributions.
If you know that you still really need it, see below on how to use the build_exe
and install_exe commands.
Some MPI-1 implementations (notably, MPICH 1) do require the actual command line arguments
to be passed at the time MPI_Init() is called. In this case, you will need to use a
re-built, MPI-enabled, Python interpreter binary executable. A basic implementation
(targeting Python 2.X) of what is required is shown below:
#include <Python.h>
#include <mpi.h>
int main(int argc, char *argv[])
{
int status, flag;
MPI_Init(&argc, &argv);
status = Py_Main(argc, argv);
MPI_Finalized(&flag);
if (!flag) MPI_Finalize();
return status;
}
The source code above is straightforward; compiling it should also be. However, the
linking step is more tricky: special flags have to be passed to the linker depending on
your platform. In order to alleviate you for such low-level details, MPI for Python
provides some pure-distutils based support to build and install an MPI-enabled Python
interpreter executable:
$ cd mpi4py-X.X.X
$ python setup.py build_exe [--mpi=<name>|--mpicc=/path/to/mpicc]
$ [sudo] python setup.py install_exe [--install-dir=$HOME/bin]
After the above steps you should have the MPI-enabled interpreter installed as
prefix/bin/pythonX.X-mpi (or $HOME/bin/pythonX.X-mpi). Assuming that prefix/bin (or
$HOME/bin) is listed on your PATH, you should be able to enter your MPI-enabled Python
interactively, for example:
$ python2.7-mpi
Python 2.7.8 (default, Nov 10 2014, 08:19:18)
[GCC 4.9.2 20141101 (Red Hat 4.9.2-1)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.executable
'/usr/bin/python2.7-mpi'
>>>
Building MPI from sources
In the list below you have some executive instructions for building some of the
open-source MPI implementations out there with support for shared/dynamic libraries on
POSIX environments.
• MPICH
$ tar -zxf mpich-X.X.X.tar.gz
$ cd mpich-X.X.X
$ ./configure --enable-shared --prefix=/usr/local/mpich
$ make
$ make install
• Open MPI
$ tar -zxf openmpi-X.X.X tar.gz
$ cd openmpi-X.X.X
$ ./configure --prefix=/usr/local/openmpi
$ make all
$ make install
• MPICH 1
$ tar -zxf mpich-X.X.X.tar.gz
$ cd mpich-X.X.X
$ ./configure --enable-sharedlib --prefix=/usr/local/mpich1
$ make
$ make install
Perhaps you will need to set the LD_LIBRARY_PATH environment variable (using export,
setenv or what applies to your system) pointing to the directory containing the MPI
libraries . In case of getting runtime linking errors when running MPI programs, the
following lines can be added to the user login shell script (.profile, .bashrc, etc.).
• MPICH
MPI_DIR=/usr/local/mpich
export LD_LIBRARY_PATH=$MPI_DIR/lib:$LD_LIBRARY_PATH
• Open MPI
MPI_DIR=/usr/local/openmpi
export LD_LIBRARY_PATH=$MPI_DIR/lib:$LD_LIBRARY_PATH
• MPICH 1
MPI_DIR=/usr/local/mpich1
export LD_LIBRARY_PATH=$MPI_DIR/lib/shared:$LD_LIBRARY_PATH:
export MPICH_USE_SHLIB=yes
WARNING:
MPICH 1 support for dynamic libraries is not completely transparent. Users should set
the environment variable MPICH_USE_SHLIB to yes in order to avoid link problems when
using the mpicc compiler wrapper.
AUTHOR
Lisandro Dalcin
COPYRIGHT
2020, Lisandro Dalcin