Can anyone comment on the number and quality of "hard core" C++ jobs and what you think the trajectory will be like? Right now I use C++ almost exclusively at work but the code/concepts involved aren't too difficult.
C++ has two big advantages over alternatives like Java: very low and deterministic processing latency (important for real-time) and very efficient memory model (important for throughput). These are really both about memory, and C++ gives you detailed control in a way few other languages do. As to why this is important, memory performance has been scaling more slowly than many other aspects of computing architecture such that it is the bottleneck for a growing number of applications. C++ allows you to be very efficient with the memory architecture without much effort. If your application is fundamentally bound by memory performance, competent C++ can get you 2-10x returns on performance in real systems relative to languages like Java. (For some other tight-loop, CPU-bound codes, not so much.)
I'm pretty bull-ish on C++11. It is not so much that I am a fan of the language but that I am a fan of what it can do in terms of performance for databases and large-scale systems. Most of the high-scale and high-performance development going forward seems to be targeted at C++ these days. That was not always the case but the requirements of modern applications are somewhat forcing that choice.
The practice of a single process locked to each core communicating over a messaging interface has trickled down to more generic massively distributed systems work because it has very good properties on modern hardware. You end up doing a fair amount of functional programming in this model because multiple tasks are managed via coroutines and other lightweight event models. This architecture is very easy to scale out because it treats every core -- on the same chip, same motherboard, or same network -- as a remote resource that has to be messaged.
MPI has one significant problem for massively parallel systems in that it has tended to be brittle when failures occur, and on sufficiently large systems failures are a routine problem. There are ways to work around it but it is not the most resilient basis for communication in extremely large systems. At the high-end of HPC MPI and similar interfaces are commonly used but for many of the next generation non-HPC systems operating on a similar scale they are using custom network processing engines built on top of IP that give more fine-grained control over network behaviors and semantics. This is not faster than MPI and often slower, and tends to be a bit more complex but it allows more robustness and resilience to be built in at a lower level. MPI was designed for a set of assumptions that work for many classic supercomputing applications but which don't match many current use cases.
also, do you have any examples of custom network engines on top of IP?
The other thing that is increasingly important in the multicore CPU space is memory locality. It's vastly more common to be limited by memory bandwidth and latency than by the execution unit. When we start analyzing approaches with a parallel complexity model based on memory movement instead of flops, the separate address space in the MPI model doesn't look so bad. The main thing that it doesn't support is cooperative cache sharing (e.g. weakly synchronized using buddy prefetch), which is becoming especially important as we get multiple threads per core.
As for fault tolerance, the MPI forum was not happy with any of the deeper proposals for MPI-3. They recognize that it's an important issue and many people think it will be a large enough change that the next standard will be MPI-4. From my perspective, the main thing I want is a partial checkpointing system by which I can perform partial restart and reattach communicators. Everything else can be handled by other libraries. My colleagues in the MPI-FT working group expect something like this to be supported in the next round, likely with preliminary implementations in the next couple years. For now, there is MPIX_Comm_group_failed(), MPIX_Comm_remote_group_failed(), and MPIX_Comm_reenable_anysource().
At least within my company we haven't used much in the way of parallel programming languages like OpenCL/CUDA/Cilk, preferring instead to use a task based concurrency library not entirely unlike Intel's TBB, but more cross platform. There's more investigation into GPGPU happening these days -- I'm curious to see if it'll pan out for more than just rendering applications. I'm also curious to see if the other (physics/audio/animation) areas can pull GPU time away from the rendering guys :)
The trajectory in this field is predictable. You get a hardware platform, have 5-7 years to figure out how to use it, and once you finally have it figured out something entirely different comes along.
Since the process was reading, process (data parallelization), writing - and reading was much bigger than writing (you read big model, animation, or texture files, and write ~8x smaller), it was "parallelized" by prefetching files in the file cache reading them in a second thread in advance.
Allowing only one writer to the HDD turned out to be beneficial (for cases like RAID SSD you can do multiple writers efficiently but this is rather pricey).
Admittedly, a lot of these jobs are oversees. However, I believe the amazon and MS consumer electronics groups are still in the US and I know for a fact that until this year all of the Toshiba TV firmwares were being done in the US.
However, my experience in the CE industry has been that we aren't worried about massively performant parallelism, just "moderately performant" parallelism. I want to say that as of a few years ago Toshiba TVs had something like 10-30 threads running at all times in parallel.