Re:Parallelism and the Limits of (programming) Languages
by btilly on Monday April 09, 2012 @02:19PM
(#3818405)
The key to writing highly parallel programs in my opinion is to have a set of primitive, highly parallelizable operations (like MapReduce) and languages that make it natural to use those operations, either implicitly or explicitly (see http://static.googleusercontent.com/external_content/untrust... for an example of such a language).
For a very different example of a more general-purpose language designed for large amounts of parallelism, see go. (Many of the same people were involved.) Note that it is much easier in that language to create accidental bottlenecks that are hard to track down than in Sawzall. (That's the trade-off when you let the programmer have more control - they can do more interesting things but also get the rope to hang themselves.)
For a very different example of a more general-purpose language designed for large amounts of parallelism, see go. (Many of the same people were involved.) Note that it is much easier in that language to create accidental bottlenecks that are hard to track down than in Sawzall. (That's the trade-off when you let the programmer have more control - they can do more interesting things but also get the rope to hang themselves.)