[FoRK] Welcome to the era of radical innovation
J. Andrew Rogers
andrew at jarbox.org
Thu Jan 9 01:12:59 PST 2014
On Jan 8, 2014, at 2:18 AM, Eugen Leitl <eugen at leitl.org> wrote:
> Christopher Willard, chief research officer at Intersect360 Research, said
> that the era of buying commercial off-the-shelf products to assemble a
> high-performance system is coming to an end. "The market should then be
> entering a new phase of experimentation, and computer architecture
> innovations," he said.
Off-the-shelf components for high-performance systems work just fine if you know how to properly design algorithms for the architecture. New computer architecture innovations won’t solve any problems.
This is all about furiously ignoring the elephant in the room of 99+% of software designers having little to no skill at designing anything but the most primitive parallel/distributed software systems. It is pretty weird that in 2014 most software engineers think largely worthless techniques like MapReduce or BSP are the pinnacle of parallel software design, and even then most can’t use them effectively for the handful of things they might be good for.
It may be a fundamental human problem. The HPC world has been in the same boat for years, and it is nominally their specialty. Mostly, HPC just papered over bad parallel software design with really expensive hardware. Fancy new architectures is the proverbial man looking under the streetlight for his lost keys.
More information about the FoRK