III. Advanced Prototype Systems 



Systems capable of scaling to 100 gigaops (billions of operations per second) performance have begun 

 to emerge. Teraops (trillions of operations per second) performance designs will be demonstrated by 

 the mid 1990s. Research in high performance systems focuses on reducing the cost and size of these 

 systems so they can be used for a broader range of applications. 



IV. Evaluation of Early Systems 



Experimental systems will be placed at sites where researchers can provide feedback to systems and 

 software designers. Performance evaluation criteria for systems and results of evaluations will be 

 made widely available. Scalability enables small to medium size systems to be used for early perfor- 

 mance evaluation and software development in preparation for larger scale applications. Larger scale 

 systems are included in the ASTA component for applications such as the Grand Challenges. 



HPCS Accomplishments 



2 Small, medium, and large scale systems developed under the HPCS component have been 

 deployed and are being used in the ASTA component. This includes large systems deployed in 

 various high performance computing centers and some systems installed in heterogeneous con- 

 figurations. 



The small and medium scale systems are being used to develop algorithms and software, includ- 

 ing fundamental building blocks for Grand Challenge problems and a wide variety of new scien- 

 tific computation models. These prototypes are characterized by very fast routing and compo- 

 nent technology, capable of scaling up to 100 gigaops system configurations. 



'^Scalable systems continue to be evaluated and refined, providing early feedback on hardware, 

 operating systems, compilers, software development tools, input/output systems, and mass stor- 

 age systems. This process has resulted in rapid upgrades in a commercial system to a scalable 

 operating system based on very small and efficient software called microkernel technology. 

 Extensions such as real time services and distributed and replicated file systems are under devel- 

 opment. 



3 New technologies are providing a scalable, modular approach to mass storage performance and 

 archiving needed in the new large scale parallel computing systems: 



- Prototype scalable mass storage systems that use parallel arrays of inexpensive disk drives 

 to achieve both high aggregate data transmission rates and large storage capacity have been 

 demonstrated. These systems demonstrate an approach that is the basis for a new genera- 

 tion of high performance file servers and mass storage systems that are internal to scalable 

 parallel computing systems. 



- Petabyte mass storage systems, which can hold images from about 30 university libraries, 

 are now available using commodity storage modules with automated robotic transfer to 

 multiple read/write units. These systems help meet the dramatically increasing require- 



30 



