Think Parallel, Think Java
For the most part, parallel programming for today's multicore and manycore architectures have been purview of C++ (think Intel's Threaded Building Blocks, Parallel Studio, RapidMind, and Cilk++) and functional languages (Erlang, Haskell, Scala, and the like). What's missing? How about Java.
That's not to say that you can't think parallel with Java. Of course you can. Java was designed from the outset for concurrent programming, with basic concurrency support in the language and class libraries. Since version 5.0, Java has included high-level concurrency APIs, such as the java.util.concurrent package, sockets, RMI, the Thread API, and message passing, among others. Plus there are plenty of third-party tools like Alan Kaminsky's Parallel Java Library and HPJava, Javelin, Titanium, and so on.
So there's no lack of parallel support in and around Java. As with other language platforms -- including C++ -- the problem continues to be complexity and the lack of abstraction. In the C++ world, this has been addressed by tools such as Intel's Threaded Building Blocks, which provides an abstraction for parallelism that avoids the low-level programming inherent in the direct use of threading packages such as p-threads or Windows threads.
Recognizing this issue in the Java world, Ateji has released Ateji PX for Java, a Java extension for providing compatibility with existing sequential code, tools, and training. The extension provides only parallelism-related aspects, with the addition of a couple of syntactic constructions.
Ateji PX is implemented by source-to-source translation to the base language, with the goal of ensuring compatibility with existing development tools. This also helps keeping the base language lean and standard. In other words, Ateji PX builds upon existing languages and tools to provide parallel programming primitives integrated at the language level.
Ateji PX is compatible with the Java language and is integrated with the Eclipse IDE. Ateji claims that it requires only minimal learning of the additional parallel constructs and no changes to the development process.
According to Ateji's Patrick Viry, "writing programs for multicore systems becomes simple, intuitive, secure and can easily be learned by all application developers." Viry goes on to say that "in just one day, one of our customers, a leading investment bank, was able to parallelize a major back-office Java application, Users were relieved when running time suddenly dropped from 40 minutes to 8 minutes."
According to the company, Ateji PX provides message-passing at the language level. This enables the compiler to map distributed programs to various target architectures and write code that is independent of any given library. Even when considering only multicore hardware, programs written in message-passing style tend to have fewer bugs (no data races, no locks), scale better (less traffic on the shared bus), and are ready for the hardware architectures of the future.
With Ateji PX, the company claims, source code written in message-passing style will run without modifications on computer clusters, MPI-based supercomputers, across a network, and in the Cloud. A distributed version of Ateji PX is in preparation, where parallel branches can be run at remote locations.