MPI Ruby
========
Project maintainer: Aaron Bedra (aaron@aaronbedra.com)
MPI Ruby is a Ruby binding of MPI. The primary goal in making this
binding was to make the power of MPI available to Ruby users in a way
that fits into the language's object oriented model. In order to do this,
the buffer and datatype management necessary in the C, C++, and Fortran
bindings have been removed. What this means is that MPI Ruby allows
you to treat objects as messages.
MPI Ruby also aims to be a complete binding to MPI in that it offers
access to nearly all functionality of MPI. While there is not a
one-to-one correspondence to functions and constants in the Ruby and
C/C++/Fortran bindings, all of the communication and topology features
are available. There are fewer methods in the Ruby binding than there
are functions in the C/C++/Fortran bindings, but this is mainly due
to the fact that the programmer no longer needs to deal with buffers
and datatypes.
Prerequesites
=============
To build MPI Ruby, you need a version of MPI installed and a version
of Ruby installed, of course. You also need to have the Ruby shared
library, libruby.so, and the Ruby header files. If these are not on
your system and you have built Ruby from source, you need to rebuild
Ruby with the --enable-shared option passed to Ruby's configure script.
If you installed Ruby from an RPM or other package, you may need to
install the 'developer package' of Ruby.
Building MPI Ruby
=================
The usual should suffice:
./configure [standard options]
make
If you want to install MPI Ruby, use this:
make install
There are three non-standard options that configure can accept as well:
--with-mpi-path=<prefix>
This option lets you specify the installation directory of your MPI
implementation.
--with-ruby-prefix=<prefix>
This option lets you specify where Ruby is installed.
--with-max-ops=<number>
By default, the user is only allowed to create 25 MPI operations. If
you would like to allow more or less, set it with this option.
Using MPI Ruby
==============
When MPI Ruby is built, an executable called 'mpi_ruby' is created.
This is an MPI program with a Ruby interpreter built in. It preloads a
Ruby module named 'MPI' to which all the MPI methods belong. In order to
run an MPI program written in Ruby with mpirun, you would use this syntax:
mpirun -np 5 mpi_ruby my-mpi-prog.rb
In this example, we're running the MPI program in the Ruby source file
'my-mpi-prog.rb' with 5 processes. You can also use any extended syntax
in this line that is supported by your MPI distribution.
A Quick Look at a Ruby MPI Program
==================================
The first thing to note about MPI programs that you will write in Ruby is that
it is no longer necessary to call MPI_Init or MPI_Finalize. Because of the
MPI runtime environment, these calls are already made for you in mpi_ruby.
Let's take a look at a program where all the processes simply identify their
ranks:
ranks.rb
--------
puts "Hello, I am #{MPI::Comm::WORLD.rank()} of #{MPI::Comm::WORLD.size()}"
--------
And to run this program, we might use this command line:
mpirun -np 5 mpi_ruby ranks.rb
And we might get this output:
Hello, I am 2 of 5
Hello, I am 0 of 5
Hello, I am 1 of 5
Hello, I am 4 of 5
Hello, I am 3 of 5
Now let's look at the program text. Notice the calls
MPI::Comm::WORLD.rank() and MPI::Comm::WORLD.size(). In both cases,
we're referencing the constant communicator WORLD in the Comm class of
the MPI module. This communicator corresponds to the MPI_COMM_WORLD
communicator in C. On this communicator, we've invoked the rank() and
size() methods that get the process' rank in the communicator and the
size of the communicator, respectively.
Let's look at an example with point-to-point communication:
basic.rb
--------
myrank = MPI::Comm::WORLD.rank()
csize = MPI::Comm::WORLD.size()
if myrank % 2 == 0 then
if myrank + 1 != csize then
hello = "Hello, I'm #{myrank}, you must be #{myrank+1}"
MPI::Comm::WORLD.send(hello, myrank + 1, 0)
end
else
msg, status = MPI::Comm::WORLD.recv(myrank - 1, 0)
puts "I'm #{myrank} and this message came from #{status.source} with tag #{status.tag}: '#{msg}'"
end
--------
Again, the invocation may look like this:
% mpirun -np 6 mpi_ruby basic.rb
I'm 1 and this message came from 0 with tag 0: 'Hello, I'm 0, you must be 1'
I'm 3 and this message came from 2 with tag 0: 'Hello, I'm 2, you must be 3'
I'm 5 and this message came from 4 with tag 0: 'Hello, I'm 4, you must be 5'
In this example, all of the processes with an even rank send a message
to the next process (unless there are an odd number of processes, in
which case the last process does nothing) and the odd ranked processes
receive a message from the previous process.
The two calls in this example that do the communication are send()
and recv() in the MPI::Comm::WORLD communicator. In the send call,
the first argument is the object to send, the second argument is the
destination process, and the third argument is the message tag with which
to send the object. If you're familiar with the C, C++, or Fortran MPI
bindings, you may notice that there is no datatype or size information
passed to send(). This is because MPI Ruby is able to treat all Ruby
objects as messages, so you needn't worry about buffers and types.
On the receiver side, the first argument to recv() is the rank of the
source process and the second argument is the message tag. recv() returns
an array of two values: the message object and the communication status.
In this example, we take advantage of Ruby's multiple assignment syntax
to store these values in msg and status. In the line after the recv()
call, note the use of the accessors status.source and status.tag.
These offer more information about the communication that just took place.
These examples should get you started and when you're ready to go further,
look at the documentation to get the syntax and semantics of more MPI
calls.
Acknowledgements
================
I'd like to thank Emil Ong for getting this terrific project started.
The new maintainer, Rudi Cilibrasi, releases his additions under the
terms of the GPL. The following copyright notice is inheritted from
the original author, Emil Ong.
I'd like to thank the very supportive Ruby mailing list members for all
their help in making this binding possible. In particular, I'd like to
thank Guy Decoux, Wesley James Landaker, Dave Thomas, Kenichi Komiya,
and, of course, matz. I'd also like to thank the entire MPICH group
here at Argonne for being so supportive of the binding.
Copyright
=========
The following is a notice of limited availability of the code, and disclaimer
which must be included in the prologue of the code and in all source listings
of the code.
Copyright Notice
+ 2001 University of Chicago
Permission is hereby granted to use, reproduce, prepare derivative works, and
to redistribute to others. This software was authored by:
Argonne National Laboratory Group
E. Ong: (630) 252-5941; e-mail: onge@mcs.anl.gov
Mathematics and Computer Science Division
Argonne National Laboratory, Argonne IL 60439
GOVERNMENT LICENSE
Portions of this material resulted from work developed under a U.S.
Government Contract and are subject to the following license: the Government
is granted for itself and others acting on its behalf a paid-up, nonexclusive,
irrevocable worldwide license in this computer software to reproduce, prepare
derivative works, and perform publicly and display publicly.
DISCLAIMER
This computer code material was prepared, in part, as an account of work
sponsored by an agency of the United States Government. Neither the United
States, nor the University of Chicago, nor any of their employees, makes any
warranty express or implied, or assumes any legal liability or responsibility
for the accuracy, completeness, or usefulness of any information, apparatus,
product, or process disclosed, or represents that its use would not infringe
privately owned rights.