A Mathematical Model for Accurately Balancing Co-Phase Effects in Simulated Multithreaded Systems

As various types of multithreaded architectures can increase resource utilization and throughput of modern processors, there exists a very large design space which must be explored to effectively characterize these systems. Since these systems execute instructions from multiple concurrent application threads, understanding the interactions between co-scheduled threads is critical to evaluating potential designs. Unfortunately, understanding the interaction of co-scheduled threads requires complex and computationally intensive simulation. Although methods for accelerating architecture simulation are effective for applications demonstrating phase behavior over time in single threaded architectures, multithreaded systems exhibit even more widely variant behavior as different phases of each thread interact. Overall, there are many simulation technology issues related to phase behavior that must be investigated to effectively explore future multithreaded architectures. This paper explores how differences in multithreaded phase interactions and their relative frequencies impact overall performance evaluation. Further, a mathematical model is presented that balances the effects of interthread phase interactions in simulation to more accurately reflect the likely behaviors encountered in a real system.