Effectively developing concurrent software for modern machines with multiple cores is perhaps the greatest technical challenge we'll encounter in some time. Our current approaches just aren't up to the task of creating robust multithreaded code.
Please see the linked post for some thoughts on this topic. Before we look directly at concurrency, let's look at what brought us to where we are.
It’s reasonably well established that categorization is a fundamental human strength. You might say that feature extraction and classification are hardware-accelerated in the brain. Even at levels far below the conscious (i.e., the visual cortex) information is categorized before it even becomes 'thought' to us. In fact, categorization is so fundamental to human thought that assuming categories themselves are real objects has been a universal illusion. In “How the Mind Works”, Pinker states that nearly all cultures initially adopt a ‘folk idealism’ as a result of this. Applying this to the boundary of man and machine communication, Object Orientation has evolved as a straightforward way to map categories (type, class, etc.) and instances of those categories onto machine architectures that deal primarily in a few primitive and largely undifferentiated numerical types.
Abstractions are complexes of categories and their interactions. Understanding complexity in terms of hierarchies of abstraction is something that people do really well. Modular Programming and, to a greater extent, Object Orientation attempt to give us tools to work at these levels of human competence - with, of course, some consequences in terms of final performance.
The predominant programming paradigms attempt to map the way people think onto the way machines operate. Let's turn our attention to concurrency and see if we can stretch this a bit further.
Concurrency as Time vs. Space
In software development, one of the things I’ve noticed is that people are much better at understanding space than time. This is why we map out time in timelines, MS Project files and a million other ways in spatial form. What this means to programming is that anywhere you can map out state in terms of space, the result is far easier to comprehend. For a clear example of this, see Google's MapReduce.
This is the core issue with concurrency. It’s very difficult to understand the possible states of concurrent systems because they happen in time. Many of the abstractions that can help with concurrency (such as Functional Programming, Message Passing, etc.) are useful because they essentially transform a state-heavy process into some equivalent but more understandable spatial map whose design appears much more static.
In some ways the game industry is at the vanguard of multicore development on consumer machines with the Xbox 360 and the PlayStation 3. The amount of time and effort that we spend finding and fixing multithreading bugs is terrifying. While the hardware companies move toward doubling the number of cores with each processor generation, software companies will be reeling.
Clearly, C++, as it is currently, is not well suited toward developing software on highly multicore machines. Ideally, we’d have a language, extension or paradigm that would allow us to map thread concurrency onto easily understood (that is, spatial) language constructs that discourage or prohibit the kind of multithreading errors that currently waste hours and hours of our time. Right now, we don’t have to worry how many registers there are on the processor when we write C++ code. Similarly, whatever language/paradigm we’d want to use would, at the compile stage, optimally generate code for the number of cores available to it on the target platform.
So would some sort of functional language be best? C++ with functional extensions? Erlang? Haskell?
What about programming models based more on hardware description languages such as VHDL and Verilog, which are inherently spatial? Would they map more effectively to multicore machinery since hardware languages describe processes which are inherently asynchronous?
In any event, we software developers have an interesting road ahead.