推荐订阅源

The Common Language Runtime has always had basic support for parallel programming in the form of locks, threads, and thread pools. With the increasing emphasis on multi-core CPUs, there has been a renewed interest in high-level parallel programming constructs.

Many people are betting on functional programming, with its emphasis on immutable objects and their inherently thread-safe properties. Haskell and the OCaml based language F# fall in this category. Others favor languages like Erlang, which are based on message passing and lightweight processes. In addition to supporting F#, Microsoft is trying several different approaches in its efforts to find the right abstraction.

The first one announced was Parallel LINQ, also known as PLINQ. Like SQL, parallelization is handled by the language itself without effort by the developer. Simply tack on an AsParallel to the query and everything else just works. Well usually, but just like SQL there are times with additional options need to be specified.

Next up is Structured Parallelism, a technique reminiscent of OpenMP. But unlike the compiler directives used in OpenMP, Microsoft's solution is based on anonymous delegates. Superficially the code changes are minor; rather than having a block of code following the "for" line, the code is passed as an argument to the "Parallel.For" function.

For code that cannot be easily expressed as queries or iterative loops, there is the Task Parallel Library. At first glance this is just another thread pool, but it supports parent-child relationships between tasks.

For those looking for finer control over their application, there is the Coordination Data Structures. These offer thread-safe collections, futures, and improved synchronization objects.