An Integrated Design Paradigm for Accelerating Full-system Simulations

In recent years, a lot of progress has been made in the field of simulator design and new simulation accleration techniques, warming technique etc have been introduced. However, despite many isolated efforts of performance enhancement, integrated efforts to benefit one proposal from other, has been lacking. In this paper, we argue for and demonstrate a unified approach for enhancing computer architecture simulations by bringing the best of all these efforts. As we show, such an integration enables full utilization of potential of different techniques and validates them over different platform. As part of our ongoing work, we demonstrate the design strategy with GEMS simulation module in simics and SMARTS sampling methodology. We also propose design optimizations for GEMS to trade speed with accuracy. Such an approach will be especially beneficial for Multi-processor cache simulations on the full-system simulator. The experiments performed over SPEC benchmarks show an average speed-up of 30 and average CPI error of around 2.4% using GEMS detailed simulator.