Effective computer architecture requires a carefully selected set of workloads that are representative of the target design's software environment. However, it is becoming increasingly difficult for computer architects to obtain workloads that accurately characterize today's complex software systems. The Architect's Workload Suite (AWS) project seeks to address this problem by developing a collection of commercial, public domain, and next-generation software systems that architects can use as a source of representative workloads. Using on-chip performance counters and custom tools, each workload in the AWS will be accompanied by Workload Characterization Worksheet that summarizes its architectural requirements including memory-system, branch-prediction, instruction-level parallelism, course-grained parallelism, and operating system support. Workload summaries will allow researchers to carefully screen the AWS workloads, selecting a set that is appropriate for their work. In addition to the workload suite, this work is developing several hardware/software analysis tools. The first, based on Trap-driven simulation, characterizes and evaluates memory systems. The second, a modified version of DECs ATOm analysis tool, generates accurate software traces of the entire system's activity, including OS and application workloads. Together, the workloads, their characterization worksheets and the tools will help architects determine how best to support future software systems.
