A methodology for performance evaluation of systems with large emulation code

Emulation affects accurate performance evaluation of applications running on the present generation of computing machines. The emulation code from present generation of machines is often passed onto the traces which when used for the trace-driven performance evaluation of the next generation of architectures results in highly incorrect evaluations. The scenario captured by the trace may be real. However, one major problem associated with such a reality is that if such traces are used for future systems design, the underlying Implicit assumption will be that the emulation problem will persist with the same degree of impact. This novel paper presents the performance evaluation problem and solutions associated with a lot of computer systems executing relatively large amount of emulation code. As a case study, we have looked at the 68k emulation code problem on Apple's Macintosh computers for today's benchmarks. We have also attempted to explore the impact of the problem and alleviate it using novel techniques.