Getting to the Bottomline in COMA Research: The Illinois Aggressive COMA Multiprocessor (I-ACOMA)

该项目的目标是探索十年后将建造的高端计算机的系统架构。 一个建议的体系结构，结合COMA和NUMA设计正在被用来探索体系结构，应用程序和编译器的问题。 研究中使用的工具包括执行驱动的模拟器，地址跟踪器，最先进的并行编译器和数据库系统，以及许多工作负载。 要探讨的体系结构问题包括合并COMA和NUMA一致性的组织、缓存标记的新用途以及一致性协议的设计。 软件问题包括用于简单COMA的操作系统内的地址转换和存储器管理、COMA中用于共享数据的数据复制的新用途、用于高速缓存一致性架构的并行代码的编译以及将并行代码中的显式数据传输编译成用于预取和转发数据的指令。

The goal of this project is to explore the systems architecture of high-end computers to be built a decade from now. A proposed architecture that combines COMA and NUMA designs is being used to explore architectural, applications, and compiler issues. Tools used in the study include an execution-driven simulator, an address tracer, state of the art parallelizing compiler and database systems, and many workloads. The architectural issues to be explored include organizations that merge COMA and NUMA coherence, new uses for cache tags, and design of coherence protocols. Software issues include address translation and memory management within the operating system for a simple COMA, new uses of the data replication that occurs for share data in COMA, compilation of parallel code for cache-coherent architectures, and compilation of explicit data transfers in parallel code into instructions for prefetching and forwarding of data.