Transcript MPJ Express: An Implementation of MPI in Java

MPJ Express: An Implementation of MPI in Java
Version: 0.35 (Beta Release)
Aamir Shafi, Bryan Carpenter, Mark Baker
[email protected], [email protected],
[email protected]
7/16/2015
1
Introduction
 MPJ Express is a message passing library that be can be used by
the application developers to develop and execute parallel Java
applications on compute clusters or network of computers.
 MPJ Express is originally designed for distributed memory
machines like clusters but also supports efficient execution of
parallel applications on desktops or laptops that contain shared
memory or multicore processors
 MPJ Express is a reference implementation of mpiJava 1.2 API,
which is an MPI-like API for Java defined by the Java Grande
Forum
 The current release contains:
– The core library
– The runtime infrastructure
– The test-suite
7/16/2015
2
MPJ Express Configurations
 MPJ Express can be configured in two ways:
– Multicore Configuration: This configuration is used by
developers who want to execute their parallel Java
applications on multicore or shared memory machines
(laptops and desktops).
– Cluster Configuration: This configuration is used by developers
who want to execute their parallel Java applications on
distributed memory platforms including clusters and network
of computers.
7/16/2015
3
Multicore Configuration
 The users can use multicore device driver (also known as shared
memory device driver) in the multicore configuration
CPU 0
Proc 0
Proc 3
CPU 3
Main Memory
CPU 1
Proc 1
CPU 2
Proc 2
Fig1: MPJ Express executing in the multicore configuration with four threads on a
quad core processor
7/16/2015
4
Advantages of Multicore Configuration
 Incremental Development:
– Users can first develop parallel applications on
desktops/laptops using multicore configuration and then take
the same code to distributed memory platforms including
clusters
 Teaching Purposes:
– This configuration is preferred for teaching purposes since
students can execute message passing code on their personal
computers
 Debugging and Profiling:
– IDEs like Eclipse can be used for easier debugging and profiling
7/16/2015
5
Cluster Configuration
 Cluster Configuration
2
– Application developers can opt to use
either of the two communication drivers
1
3
in the cluster configuration
– Java NIO Device Driver (niodev) It can be
4
used to execute MPJ Express programs
0
on Ethernet-based interconnects.
5
– Myrinet Device Driver (mxdev): Many
clusters today are equipped with high
performance low latency networks like
7
Myrinet.
– Lets consider a cluster comprising of 8
6
compute nodes.
Fig2: MPJ Express executing in the cluster
– Every process is started on different
configuration with 8 processes on a
node while running code with Cluster cluster with 8 compute nodes
Configuration, depending upon the
machines file
7/16/2015
6
MPJ Express Installation (UNIX/Linux/Mac)
 Pre-requisites
– Java 1.5 (stable) or higher
– Apache ant 1.6.2 or higher (Optional)
– Perl (Optional)
 Running MPJ Express Programs in the Multicore Configuration
– Download MPJ Express and unpack it.
– Set MPJ_HOME and PATH environmental variables:
• export MPJ_HOME=/path/to/mpj/
• export PATH=$PATH:$MPJ_HOME/bin
–
–
(These above two lines can be added to ~/.bashrc)
Write your MPJ Express program (HelloWorld.java) and save it.
Compile: javac -cp.:$MPJ_HOME/lib/mpj.jar
HelloWorld.java
–
Execute: mpjrun.sh -np 4 HelloWorld.java
7/16/2015
7
MPJ Express Installation (UNIX/Linux/Mac) Cont’d
 Running MPJ Express Programs in the Cluster Configuration
– Download MPJ Express and unpack it.
– Set MPJ_HOME and PATH environmental variables:
• export MPJ_HOME=/path/to/mpj/
• export PATH=$PATH:$MPJ_HOME/bin
–
–
–
–
–
–
(These above two lines can be added to ~/.bashrc)
Write a machines file (name it “machines”) stating hostnames or IP
addresses of all the machines involved in the parallel execution
Write your MPJ Express program (HelloWorld.java) and save it.
Start Daemons: mpjboot machines
Compile: javac -cp.:$MPJ_HOME/lib/mpj.jar HelloWorld.java
Execute: mpjrun.sh -np 4 HelloWorld.java
Stop daemons: mpjhalt machines
7/16/2015
8
MPJ Express Installation (Windows)
 Pre-requisites
– Java 1.5 (stable) or higher
– Apache ant 1.6.2 or higher (Optional)
– Perl (Optional)
 Running MPJ Express Programs in the Multicore Configuration
– 1. Download MPJ Express and unpack it.
– 2. Set MPJ_HOME and PATH environmental variables. Right-click My
Computer->Properties->Advanced tab->Environment Variables and
export the following system variables (User variables are not enough)
• Set the value of MPJ_HOME = c:\mpj (assuming mpj is in c:\
• Append the c:\mpj\bin directory to the PATH variable
– Cygwin on Windows: (assuming mpj is 'c:\mpj‘): The recommended way to
is to set variables as in Windows. If you want to set variables in cygwin
shell
• export MPJ_HOME="c:\\mpj“
• export PATH=$PATH:"$MPJ_HOME\\bin"
7/16/2015
9
MPJ Express Installation (Windows) Cont’d
– Write your MPJ Express program (HelloWorld.java) and save it.
– Compile: javac -cp.:$MPJ_HOME/lib/mpj.jar HelloWorld.java
– Execute: mpjrun.bat -np 4 HelloWorld.java
 For running MPJ Express programs in the cluster configuration, refer to the
$MPJ_HOME/README-win.txt and $MPJ_HOME/doc/windowsguide.pdf
7/16/2015
10
Writing Programs with MPJ Express
 A sample Program with MPJ Express is
import mpi.*;
public class HelloWorld {
public static void main(String args[]) throws Exception {
MPI.Init(args);
int me = MPI.COMM_WORLD.Rank();
int size = MPI.COMM_WORLD.Size();
System.out.println("Hi from <"+me+">");
MPI.Finalize();
}
}
 Save this Program as HelloWorld.java
7/16/2015
11
Executing MPJ Express Programs
 Compiling
– UNIX/Linux/Mac:
javac –cp .:$MPJ_HOME/lib/mpj.jar
HelloWorld.java
– Windows: javac –cp .:%MPJ_HOME%/lib/mpj.jar HelloWorld.java
 Running with Multicore Configuration
– UNIX/Linux/Mac: mpjrun.sh –np <no. of processors>
HelloWorld.java
– Windows: mpjrun.bat –np <no.
 Running with Cluster Configuration
– Start the daemons
– UNIX/Linux/Mac: mpjrun.sh –np
of processors> HelloWorld.java
<no. of processors> -dev niodev
HelloWorld
– Windows: mpjrun.bat
–np <no. of processors> -dev niodev
HelloWorld
7/16/2015
12
Executing Programs with MPJ Express on UNIX/Linux/Mac
 Multicore Configuration
 Cluster Configuration
7/16/2015
13
Executing Programs with MPJ Express on Windows
 Multicore Configuration
 Cluster Configuration
7/16/2015
14
Additional Runtime Arguments
 The mpjrun.[sh/bat] script accepts some additional arguments which can be
passed according to the user preferences. Some of them are
– np switch: If a value is specified, then MPJ Express starts those many
processes for the user application. It can be specified like
mpjrun.[sh/bat] –np 2 ProgramName.
– dev switch: This switch can be given at runtime to change to the
appropriate device driver. It can be used as mpjrun.[sh/bat] –dev
multicore/niodev/mxdev ProgramName.
– machinesfile switch: While running programs, MPJ Express runtime
expects a file with name “machines” in the current directory. User can also
specify a machines file with different name using machinesfile switch as
mpjrun.[sh/bat] machinesfile myFile ProgramName
– wdir switch: If your programs reads some file, then it may be a good idea
to separate this file from your application classes, or copy it to a tmp
directory and specify this tmp directory as working directory using -wdir
switch.
7/16/2015
15
Manuals and Help Resources
Windows
Linux
README (Quick
Start Guide)
$MPJ_HOME/README-win.txt
$MPJ_HOME/README
User Guide
$MPJ_HOME/doc/windowsguide.pdf
$MPJ_HOME/doc/linuxguide.pdf
 Visit MPJ Express (http://mpj-express.org) for more details.
7/16/2015
16
Contact and Support
 MPJ Express Users Mailing List
(http://www.lists.rdg.ac.uk/mailman/listinfo/mpj-user)
 Alternatively, the users can contact us directly by email.
– Aamir Shafi ([email protected])
– Bryan Carpenter ([email protected])
– Mark Baker ([email protected])
7/16/2015
17
Contributors
Aamir Shafi
Jawad Manzoor
Kamran Hamid
Mohsan Jameel
Bryan Carpenter
Mark Baker
Hong Ong
Guillermo Taboada
Sabela Ramos
7/16/2015
18
Acknowledgements
 The project partners received a generous grant from the British
Council under the PMI2 Connect program
(http://www.britishcouncil.org/learning-pmi2-connect.htm)
 Project Title: CollAborative Multicore Programming Using
Scientific Java Messaging


Project Start Date: September 2008
Duration: 2 Years
7/16/2015
19