The aggregate networked compute power in the United States today is estimated to be between 10 and 100 TeraFLOPS; by the end of the decade it is expected to reach a PetaFLOPS. The focus of this workshop was to determine the means of implementing a single computer capable of more performance than all the computers on Earth put together. Far from being treated as unduly premature or even superfluous, the goal was dealt with seriously and in earnest by all participants. Why?
The fundamental issue driving this workshop is that the stakes are too high for the U.S. to leave this future technology in the hands of other nations. It may be that dealing with the PetaFLOPS question now will not have a demonstrable effect on the computing industry for some years. But, when the technology begins to have impact, the ground work will have been laid years in advance, and only those nations having previously invested in the underlying enabling technologies will be positioned to play a commanding role in its exploitation.
Results from this workshop indicate that the U.S. is already lagging in specific technologies that may be crucial to PetaFLOPS computing. The U.S. is no longer the premier producer of high-density memory technology. Yet, one of the key findings is that memory will be the major component determining both the capability and the cost of such systems. A nation that cannot produce competitive high-performance memory will not be able to compete in the PetaFLOPS market. Another example is cryogenic Josephson Junction technology. Once the world leader, the U.S. has been displaced by Japan in the production and application of this technology and now we have seen that it may be crucial to low power, high-speed processing at the PetaFLOPS level.
This workshop constituted a deliberate and concerted effort by the U.S. to determine the critical capabilities required to take a leadership role and to establish now an initial path in that direction. An open issue at the beginning of this workshop was the merits of the subject itself. The implications of the workshop findings is that not only does the topic warrant consideration and is timely, but that early action must be taken in the areas of devices, architecture, and system software to ensure U.S. capability into the next century. Therefore, an important implication of the findings of the workshop is the legitimization of the discipline of PetaFLOPS computing systems.