Developing the architecture of a computer is a tricky business at best; one has to consider not only present programs, but possible future programs as well. There are simulations to assist the architect, but these take days, and sometimes weeks, to simulate minutes of the program on high performance computers.
The authors propose that, by exploiting repetition in common applications, simulation time could be reduced to hours, and possibly minutes. The authors use a simulation called EXPERT to demonstrate their proposal. This speed improvement was achieved with little loss in accuracy, specifically, the loss was found to be on the order of one percent or less.
The paper is organized into seven sections, beginning with an introduction. The authors follow the introduction with some observations, the methodology used to exploit repetition, the experimental setup used, the evaluation and analysis, related work, and, finally, conclusions. After reading the paper, I was impressed by the ease with which I could read it, and get insights from it. I believe that this paper is a useful one, which should be read and kept for future reference.