The blog got me thinking …
I agree with the taxonomy – may be Category 2 should be called sequential-parallel programming or even sequentially programmed – parallel via automation !
The Category 3 – Inherently parallel programming, IMHO, should deal with massively parallel architecture – for the current generation (i.e. < next 2 years!) it would mean ~ 32 cores/threads. For example Sun’s Niagra1 has 8 cores – each with 4 threads crossbar-ed to 8 L2 cache. That is where parallel programming makes sense and impact. Arch is correct in talking about the cases where the application threads outnumber the hardware threads. But in case where the chip has massively parallel architecture, the reverse will happen – i.e. software with a few threads and the hardware threads outnumber them, the performance will suffer … In short, look at inherent parallelism for the massively parallel architectures …
