ISC HPC Blog
Is the Killer App Dead?
Was it a victim of its own success?
HPC is a tool. We use it to solve problems and make discoveries. At the highest end of HPC, it’s all about capability, not capacity. How does one demonstrate or sell a new capability? Since its inception, the approved solution to this problem has been the killer app.
In the beginning, we didn’t call them killer apps. They were the “Grand Challenges.” The first collection of grand challenges was described in February of 1991 when the US government’s Office of Science and Technology Policy released the first Blue Book – a supplement to the President’s FY 1992 Budget Request for the newly created High Performance Computing and Communications Program. The Blue Book was entitled “Grand Challenges: High Performance Computing and Communications” and contained a listing of the computational science and engineering challenges seen as drivers for federal expenditures on HPC at that time.
As I pointed out in an HPCwire article a couple of years ago (Meet the Exascale Apps, 12 April 2012), those apps haven’t changed much in the past twenty years, and, with few exceptions, they are the current set in global use.
Are the killer apps working for us? Some observers think not. The argument has been made that, as HPC has successfully diffused through many application disciplines over the past decades, the killer apps have morphed into what might better be called the “usual suspects.” So, given that exascale computing projects are currently being funded on several continents, how were they justified? And what does this portend for the future of HPC?
It appears that the killer app justification is being superseded by a computer technology one. That is to say, the new model uses the next-generation computer as its own justification. The more elaborate form of this new argument invokes both opportunity and peril. The opportunity is in technology trickledown to larger markets for computing equipment. The peril part appeals to national and economic security concerns about what might happen if some competitor or adversary were to get so far ahead in HPC that catching up became exceedingly difficult.
If this new justification is working – and it certainly appears to be – are killer apps simply lacking, or have they been made irrelevant by their earlier successes? Does technology push now win out over application pull? In any case, should we care?
In that previously mentioned HPCwire article, I did suggest a few new application candidates to refresh the killer category. So far, the only one that appears to have good traction is Henry Markram’s Human Brain Project. Big Data and the Internet of Things may be sources of future killer apps, but right now they don’t seem to be pushing the computing capability envelope. Although most HPC centers are loath to admit it, Linpack (the app that’s not really an app) remains a killer app in the sense that it certainly plays a large role in justifying high-end systems. But, in its current form, even Linpack may fall by the wayside as we move to exascale systems.
Ansatz: Technology push is winning and, if not dead, the killer app is pretty moribund.
Should we care? Yes
In general, as HPC grows in capability, it can be used in three distinct modes:
To my way of thinking, the third mode is by far the most interesting and the one holding the greatest promise. It is also where applications are born. Since our current killer apps seem to be dying off, perhaps we should focus more on bringing new ones into the world.
If you think I’ve misunderstood either the plight of killer apps or the persistence of the role they should play in HPC, please feel free to let me know what you think.