InfoQ catches up with Gary Frost from AMD who unveiled an alpha release of Aparapi, an API that allows programmers to write logic in Java to be executed on a GPU. GPUs are the massively parallel hardware acceleration chips originally installed in PCs to boost graphics rendering performance but that are now pushed to other kinds of compute-intensive tasks that have nothing to do with graphics.
The patterns&practices team has released Parallel Programming with Microsoft .NET, a book containing guidance for writing parallel programs for .NET. In essence the book contains 6 design patterns for parallel programming accompanied by code samples.
Optional parameters have always been part of .NET, but with C# unwilling to support it, using them was generally considered taboo unless work with COM libraries. Now that C# 4 does support them, we are starting to see them used for a lot more than just legacy code. Other uses include interoperability with dynamic languages, immutable data structures, and various parts of ASP.NET MVC.
Brahma is an open source C# library that provides support for parallel computations running on a variety of processors. Currently, Brahma has a GPU provider but its modular structure allows using different providers for other types of processors. One C# method can contain both statements running on CPU and GPU without additional glue code.
Even though Microsoft has been working on .NET’s Parallel Extensions since 2007, there are still many features that they didn’t have time to fully implement for .NET 4.0. Some features were “too application-specific to be included in the core of the Framework” while others simply needed for testing and user feedback. So instead they are being released as a set of patterns and samples.
Accelerator V2, currently a preview build, is a .NET managed library easing the task of writing data-parallel programs executed on multi-core CPUs and GPUs.
FlightCaster recently open sourced Crane, a tool for distributing and remotely controlling Clojure instances, currently specialized for EC2. Incanter is a Clojure library and tool that makes R-like statistical computations easy with Clojure. Also: the build and dependency management tool Leiningen 1.0 is now available.
Microsoft Biology Foundation is a collection of libraries build on the .NET framework and based on traditional open source traditions. Rather than reinvent the wheel, Microsoft is leveraging the file formats already found in bioinformatics community. Even more unusual for them, they are soliciting contributions to be added to future versions of MBF.
Go is a Google experimental open source new language resembling C but adding features like reflection, garbage collector, dynamic types, concurrency, and parallelism.
Task Parallel Library, .NET 4.0’s replacement for ThreadPool, got a face lift for beta 2. In addition to performance improvements, it The most important change is probably the new cancellation framework that replaces parent/child relationships with cancellation tokens that can be freely given to logical groups of tasks.
.NET 4 will have new types to support building cancellation-aware applications and libraries. The new CancellationToken, CancellationTokenSource, and cancellation exception types provide a cooperative cancellation framework.
In his keynote at Sapphire 2009, Hasso Plattner, co-founder of SAP, explained that he sees multi-cores, and in memory column-oriented databases in the future of enterprise applications.
InfoQ announced Microsoft’s plan to ship Axum, an incubation language project, 2 weeks ago. In the meantime Microsoft has finished packaging an early release (v. 0.1) and made it available for download.
Axum, previously known as Maestro, is a Microsoft incubation language project meant to provide a parallel programming model for .NET through isolation, actors and message passing. The language borrows many concepts from Erlang but with a C#-like syntax.
In an essay on Cilk Arts, Professor Guy Blelloch argues that parallel programming is not intrinsically hard, but rather a question of abstraction. The three problems identified by Blelloch are a lack of training in parallel thinking, separating parallel implementation from algorithms, and determinism. After detailing each, he explains why he thinks they can be overcome.