List of research topics on Java for HPC

1. Java as a Language for HPC
2. Java Compilation and Optimization for HPC
3. Run-time Support to Java for HPC
4. Miscellaneous

• Java as a Language for HPC

This area contains research proposals of Java extensions to support high-performance computing. This includes support to parallelism (multidimensional arrays, parallel loops, communication), and improvements to numerical precission. In this area there are also some efforts in highly optimized implementation of numerical libraries.
 
 
• High Performance Java

Definition of a Java extension to support data parallel programming. The basic extensions include distributed arrays, data parallel distributed loops, and a set of communication primitives [PostScript].

Basicament el que defineixen es una extensio al Java per a donar suport a operacions amb arrays multidimensionals i distribuits (data parallel programming). La sintaxi que utilitzen sembla intuitiva per algu que no ha programat mai amb paral.lelisme de dades; pero resulta molt empalagosa si estas acostumat a llenguatges tals com Fortran-90, HPF, and so on.
 

• Research group: NPAC at Syracuse University
 
• Experiments with HPJava

Several experiments are performed to evaluate Java as a natural candidate for portable parallel programming language. These cover socket programming, message-passing (MPI), and class libraries for data-parallel programming [expHPJ]. They also provide and evaluate a programming model (extending Java, Fortran, and C++) that is a flexible hybrid of the data-parallel, language-oriented, HPF style, and SPMD style. This is refered as the HPspmd model [hpspmd]. Some other discussions and proposals [java4parprog, java4simulat].
 
• Research group: NPAC at Syracuse University
 
• Level 1 BLAS in Java

They propose to extend Java with some mathematical primitives to improve the performance of certain operations while mantaining portability. On parallel systems, this is achieved by exploiting the multi-threading mechanism [PostScript].
 
• Research group: Indiana University
 
• Titanium: A High Performance Java Dialect

Titanium is a languaje and system for high-performance parallel scientific computing. Titanium is Java with support to multi-dimensional arrays, and programs are explicit parallel SPMD with a global address space [PostScript].
 
• Research group: Lawrence Berkeley National Laboratory
 
• Spar: A Language for Semi-automatic Parallel Compilation

Spar is a programming language for semi-automatic parallel programming, in particular for array-based applications. Spar is a superset of Java, extending it with constructs for parallel programming and support for array manipulation [PostScript].
 
• Research group: Delft University of Technology
 
• High Performance Parallel Programming with Java Libraries

They propose to exploit massive parallelism by binding native standard libraries to Java. These libraries include: MPI, BLAS, BLACS, PBLAS and ScaLAPACK [PostScript]. JavaMPI is their effort to bind Java programs to the native MPI library for running on distributed-memory machines.
 
• Research groups: University of Westminster and IBM T. J. Watson Research Center
 
• SPMD Programming in Java

They analyze the suitability of Java concurrent constructs to write high-performance SPMD code for distributed-memory parallel machines. Their conclusion is that efficient implementations are feasible, although they propose a library of Java methods to ease SPMD programming [ibm_spmd].
 
• Research group: IBM T. J. Watson Research Center
 
• Java Language for HPC

In their study they discuss some techniques for using Java in scientific and engineering applications. They review some relevant features of the language and study how Java-based developments can help scientific application developers [PostScript].
 
• Research group: Boston University and Cooperating Systems Corporation
 
• Java as a Front-End to HPC

This is a study of Java as a front-end language for high-performance systems running native code. Severeal approaches are analyzed to provide powerful back-end services to Java client programs (remote shell, PRC, RMI, network access mechanism from Java, and some other research projects) [PostScript].
 
• Research group: University of Adelaide
 
• Other proposals: Java for (non-parallel) numerical applications

Includes some proposals that cope with numerical precission, and numerical libraries.
 
• Java for High Performance Numerical Computation

A set of proposals to extend Java for high performance numerical computing. Basically the extensions deal with numerical precision, complex numbers, and support to multi-dimensional matrices. JavaSoft
 
• Numerical Libraries with Java

They address practical issues of the Java language and environment which have an effect on numerical library design [PostScript]. NIST Gaithersburg, University of Tennessee, Oak Ridge National Laboratory, and University of Maryland
 
• A Matrix Math Library for Java

They define a framework based on numerical algorithms operating on matrices available for use by all the semiconductor toolsets [PostScript]. MIT Microsystems Technology Laboratories
 
 
 
• Java Compilation and Optimization for HPC

This area contains a list of different compiler optimizations for Java, including restructuring Java to Java compilers, bytecode optimizers, and just-in-time compilers. All these proposals are targeted to high-performance computing (parallel systems or numerical applications).

 

• javar: Restructuring Compiler for Java

Restructuring compiler for Java, which can be used to make implicit parallelism in a Java program by means of multi-threading. They detect implicit parallelism from loops, or let the programmer to identify all implicit parallelism by means of annotations [PostScript].
 
• Research group: Indiana University
 
• javab: Bytecode Parallelization

Techniques for the automatic detection and exploitation of implicit loop parallelism in bytecode [ref]. Implicit parallelism is made explicit by means of the multi-threading mechanism of Java.
 
• Research group: Indiana University
 
• Briki: Optimizing Java Bytecodes

They define a high-level intermediate representation for Java (JavaIR), and propose some optimization techniques which are only possible or much easier to perform using JavaIR. They show trade-offs between optimizing Java, and other machine-dependent optimizations that can only be performed from bytecode [PostScript].
 
• Research group: University of Rochester
 
• Briki: Just-in-Time Java Compilation

Proposal of just-in-time optimizations techniques for high-performance Java programs, based on faster analysis during the optimization stage [PostScript].
 
• Research group: University of Rochester
 
• Just-in-time Compilation (II)...

Paper from PLDI'98...
Fast, effective code generation in a just-in-time Java compiler
Ali-Reza Adl-Tabatabai, Michał Cierniak, Guei-Yuan Lueh, Vishesh M. Parikh and James M. Stichnoth
 
• Research group: Intel
   
• Uniform IR for Parallel Compiler Research

A compiler construction framework is described. The system analyzes the input source language and converts it into a uniform intermediate representation that can be latter manipulated and optimized.
 
• Research group: NPAC at Syracuse University
 
• Optimizing Java bytecode

Aggressive optimizing compiler for Java bytecode. Some classical optimization strategies are harder to implement due to the Java language design, therefore they exploit some new optimizations unique for this language [PostScript].
 
• Research group: Rice University
 
• Bytecode Annotation

They propose an optimization approach based on bytecode annotations during the Java to bytecode translation, allowing both traditional interpretation by a JVM, and aggressive optimizations by a bytecode compiler [PostScript].
 
• Research group: University of California at Irvine
 
• Other Proposals for (non-parallel) Java Optimization

 
• IBM High Performance Compiler for Java

The High Performance Compiler for Java takes the bytecode and converts it into native code for the target system, removing the JVM from the loop. IBM T. J. Watson Research Center
 
• CACAO: A 64 bit Just-in-Time Compiler

The CACAO system translates Java bytecode into native code for the Alpha processor. Lacal variables and stack locations are replaced by pseudo-registers eliminating the 32 bit restriction on address types [PostScript]. Technische Universitat Wien
 
• Crearn: Bytecode Optimization

Crearn is an optimizer compier for Java bytecode, using side-effect analysis to improve optimizations. They implement dead-code elimination and loop invariant removal [PostScript]. University of Aarhus
 
• Java Optimization Tips

Some tips to optimize Java programs in terms of speed, size and maintanability. Also includes a summary of some features of modern compilers. Carnegie Mellon University (no longer maintained).
 
• Fortran to Java Converter

f2j is a prototype of a Fortran-77 to Java converter. Translation issues are identified, and several approaches are presented [PostScript]. NPAC at Syracuse University
 
• Fortran to Java Converter

This other group has also developed a Fortran to Java translator (f2j) to convert Fortran numerical subroutines into class files suitable for use by Java programmers [PostScript]. In the paper, they also discuss an environment to access parallel resources through the net. University of Tennessee and Oak Ridge National Laboratories
 
 
 
• Run-time Support to Java for HPC

This area contains run-time systems to support the execution of Java applications on parallel systems. This includes run-time parallel libraries, distributed implementations of JVM, and heterogeneous parallel environments for Java. Compilation issues at run-time (jit compilersr) have been included in the Java compilation area.
 
 
• Run-time Kernel Library

They define a high-level run-time library for data-parallel languages. The goal is to provide a coherent interface to the distributed array descriptors and collective operations needed for an efficient translation of parallel constructs. The kernel is currently implemented on top of MPI, and the library design is object-oriented.
 
• Research group: NPAC at Syracuse University
 
• JMPI: MPI Support for Java

JMPI is a commercial effort to develop a message-passing framework and parallel support environment for Java. They deal with aspects of the programming model (task, data, and data-flow parallelism) as well as the parallel environment.
 
• Research group: MPI Software Technology
  
• MultiJav: DSM System for Multiple JVM

They propose a distributed shared memory system as a distributed implementation of the Java Virtual Machine (JVM). Their goal is to provide an efficient, portable, and distributed shared memory system for flexible programming based on Java [utah].
 
• Research group: Utah State University
 
• Support of Parallel Sparse Computation in Java

They describe a run-time environment with continuous compilation that supports automatic parallelization of sparse computations on distributed environments. Programs are written using high-level matrix notations in Java and matrix notations are rewritten into parallel operations at run-time (using profiling information and according to a cost model) [PostScript].
 
• Research groups: National Tsing-Hua University and Academia Sinica
 
• Visper: An environment for Network Parallel Programs

Visper is a visual environment to develop and run parallel programs on NOW, based on the MPI standard.  They also propose a parallel extension to Java to handle remote processes and group communication [PostScript].
 
• Research group: Macquaire University
 
• Heterogeneous Parallel Environments for Java

There are multiple research groups developing environments that support parallel execution on heterogeneuos systems. Some of them are run-time systems, and some others are based on a set of parallel libraries that can be called from Java programs.
 
• VPL: A Web-based Programming Environment

The Virtual Programming Environment is a Web-based virtual programming environment that can be accessed on any platform using a standard Java-enabled browser. VPL facilitates the development and execution of parallel programs [PostScript]. NPAC at Syracuse University
 
• Nexus: Technologies for Ubiquitous Supercomputing

NexusJava is a communication library that enhances the Java programming language with global pointer and remote service request mechanisms. Nexus allows the implementation of advanced languages, libraries, and applications in heterogeneous parallel distributed environments [PostScript]. Argone National Laboratory
 
• Javelin: Internet-based Parallel Computing

Javelin is a Java-based infrastructure for global computing that runs on a heterogeneous network of several parallel machines, workstations, and PCs [PostScript]. UC Santa Barbara
 
• Java/DSM: A Platform for Heterogeneous Computing

They describe a system that combines Java and DSM systems, to transparently handle data communication between machines and eliminate the need for the programmer to write message-passing code [PostScript]. Rice University
 
• IceT: Distributed Computing with Java

The IceT project is a framework for collaborative and high-performance distributed computing with Java [PostScript]. Emory University
 
• Jet: Web-based Metacomputing

Jet is a parallel library implemented on Java that supports the execution of parallel applications over machines that are connected to the Web [PostScript]. Universidade de Coimbra
 
• Web-based Framework for Parallel Computing

They present JAVADC, a Web-Java based framework to enable parallel applications written using PVM, PPVM, and MPI [PostScript]. Old Dominion University and ICASE, NASA Langley Research Center
 
• NetSolve: Java Access to Numerical Libraries

NetSolve allows users to access pre-installed computational resources distributed across the network, so the user can perform scientific computing without having any computing resource in his computer. In addition, they have developed a Fortran to Java translator (f2j) to convert Fortran numerical subroutines into class files suitable for use by Java programmers [PostScript]. University of Tennessee and Oak Ridge National Laboratories
 
 
 
• Miscellaneous

Some discussions and studies related to Java and its impact in high-performance computing.
 
• Web/Java environment for HPC and Computing on the Web

Some discussions about the recent developments on the Internet and their implications for high-performance distributed computing. Petaop and Exaop infrastructure of parallel processor environments will offer new opportunities for high-performance simulations. Also basic web technologies are discussed and some real examples are provided in [PostScript]. NPAC at Syracuse University
 
• A Study of Java in Technical Computing Markets

Java can be orders of magnitude slower that optimized machine code. In this study they quantify the performance of interpreted bytecode in numerically intensive technical applications [PostScript]. Sun Microsystems
 
 

• under construction: other related links...
• Java conferences
• Java Grande and Java Forcse (related information)

Last modified 14/08/98
by Jordi Garcia