New Language on the Block: Java for High-Performance Computing?
by Mike Ashworth
The Java language is not only an emerging new technology in Web-based
computing, but is starting to have a considerable influence in
many IT-based science and engineering application areas. For example,
it is emerging as a serious contender for use in High Performance
Computing (HPC). At the CLRC Daresbury Laboratory, we are developing
a multi-language approach in which Java is used as a front-end
to existing HPC programming environments. This allows the utilization
of high-performance legacy Fortran and C codes within a Java wrapper
that facilitates the construction of Graphical User Interfaces
(GUI) and access to Web-based client-server computing.
The Java language was designed by James Gosling at Sun Microsystems
starting in 1990. It was first intended as a new language for
embedded systems in consumer microelectronics, which led to the
basic design features of Java: it is simple, object-oriented,
architecture neutral, robust, secure and extensible. In 1993,
as the World Wide Web was developing to a more graphically oriented
interface, the Java team realised that Java also had ideal qualities
for Web-based applications. Java has some similarities with C++,
but it is much simpler. Many of the more advanced features of
object-oriented programming, such as operator overloading, pointer
arithmetic and multi-dimensional arrays, were left out.
How does Java work?
Java source is compiled to class files that contain machine-independent
byte-code. This is like machine code in form but is not specific
to any particular hardware. When a Java enabled client, such as
Netscape or Internet Explorer, accesses a page containing a Java
applet (a small Java application embedded in a Web page), the
byte-code is downloaded and runs on the client’s own hardware
using an interpreter known as the Java Virtual Machine. Because
Java is compiled to byte-code, it can run on any platform for
which an interpreter has been written. Java is small so it can
run efficiently on anything from PCs upwards, and the interpreter
takes only a few hundred kilobytes.
Java and High Performance Computing
Java is more efficient than other scripts, but it is still very
slow compared to traditional languages. Even if compiled to native
machine code, there are features in the language that make it
currently impossible for native Java compilers to apply the normal
optimizations which allow Fortran and C programs fully to exploit
the hardware of state-of-the-art. For example, security considerations
in Java demand that all array references are checked for out-of-bounds
indexes at run-time, which effectively kills performance for computational
loop kernels. Experiments at IBM have shown that if this checking
can be performed at compile-time and the compiler is allowed to
make other optimizations, Java code improves from 1% of Fortran
to 80-100%.
Vendors, software developers and computer users in the US have
formed a group known as Java Grande, which believes that Java
has the potential to be a better environment for HPC than any
previous language. The Java Grande Forum aims to develop a consensus
for the use of Java for Grande applications and to make recommendations
for changes to the Java standard and for the establishment of
standards for libraries and services. Interest is also growing
in the UK. The First UK Workshop on Java for High Performance
Network Computing was held at Euro-Par ‘98 in Southampton in September
1998. There are currently moves to form a UK Java Grande Forum
similar in style to the US Forum, but with a greater emphasis
on the services and tools required by high performance applications.
Java, HPC and the Web
The World Wide Web was conceived as a means of accessing information,
but it is increasingly evolving towards a means of providing services.
A major sea change is happening on the Internet, due in no small
part to the emergence of the Java programming language. Java is
expanding rapidly and comes complete with a wide range of class
libraries and toolkits for Web access, thread based computing
and GUIs.
Although the popular image of the Web has been as a global information
provider, this has not been its only manifestation. Increasingly,
organisations are setting up an Intranet - a corporate Web server
providing internal information and services. In an industrial
R&D environment, the Intranet may enable access for everyone in
a department to the technical designs and specifications of a
new product together with the project specifications, modification
records, test procedures, test results, marketing information
etc.
It is in this environment that we see the major benefits of integrating
Java, with its support for Web-based computing, with traditional
high-performance programming using Fortran and C. This vision
is of a new type of multi-language multi-paradigm programming
environment for high performance applications which makes use
of innovative Web-based and GUI technologies provided by Java
and provides a high performance and simple API and a high level
of portability and flexibility for large-scale applications.
It emphasises the strengths of each component: Java allows easy
access to Web-based computing and facilitates the construction
of GUIs; Fortran and C allow large-scale applications best to
exploit high performance hardware.
Please contact:
Mike Ashworth - CLRC
Tel: +44 1925 60 3663
E-mail: M.Ashworth@dl.ac.uk
