Performance Evaluation of Multithreaded Systems

This paper introduces a stochastic model to evaluate memory access performance in multi-threaded systems, where multiple threads access a shared memory segment independently. The model posits that memory access times are governed by a Poisson process with exponentially distributed intervals between accesses. We derive and analyze several key performance metrics, including the probability of no contention, the probability that multiple threads will contend for access simultaneously, and the expected delay in accessing memory. The findings demonstrate that as the number of threads increases, contention rises significantly, resulting in marked performance deterioration. Numerical evaluations further underscore the exponential growth in contention and memory access delays as more threads are introduced. The results emphasizes the critical role of effective synchronization mechanisms to ensure scalability in multithreaded systems.