A Systems Approach to a Modern Computer Architecture Laboratory

Computer architecture has been typically taught from a CPU-centric, depth-over-breadth, context-free perspective. The efficacy of this approach for computer science and software- engineering majors seems questionable. We will develop a laboratory curriculum for computer architecture which places the ideas of computer architecture in a systems context, emphasizing applications where knowledge of architecture is essential for software and system development. The areas on which we will focus our new computer systems architecture course are architectural subsystems (CPU, memory, I/O), performance evaluation, parallel architectures, and high performance computing. We will use simulations and experiments as a means of teaching concepts, as well as important activities in themselves. We will develop fourteen laboratory exercises. A shared-memory multiprocessor will enable the study of modern architectural subsystems, support experimental performance evaluation including benchmarking, and teach an introduction to software development on a commercially common multiprocessor. Other labs will use performance monitoring and simulation software to study architectural subsystems, their interaction, and their impact on performance. Commercial software to support the network of workstations computing model will give students insights into parallel processing. A short and basic introduction to logic gates and circuits with accompanying breadboard laboratory will de-mystify the computing elements of modern computers. The evaluation plan for our project includes a longitudinal study to assess student change in attitude towards computer architecture, and a second study to measure the efficacy of our laboratory based curriculum on student's knowledge of fundamental computer architecture concepts.