HPC architecture requirements for earth and space science applications

Scalable high performance computer architecture brings substantial processing, memory, and communications resources to large scientific and engineering applications while leveraging industrial investment and rapid growth in VLSI microprocessor technology. These processor architectures, while approaching supercomputer performance, were developed to support the workstation market and incorporate policies consistent with the needs of that user community. Large parallel computers differ from workstations in that they exhibit high diameter and experience extremes in memory access latency. These architectures employ sophisticated cache structures to minimize average latency and achieve high apparent memory bandwidth. Unfortunately, many large scientific codes do not exhibit memory access patterns with the necessary locality of reference on which effective use of cache mechanisms relies. This presentation examines the memory access patterns of applications from the Earth and space sciences community and considers their implications for cache based architecture.