Performance Evaluation of Multithreaded Systems

EasyChair Preprint 15279

Abstract

This paper presents a stochastic model-based evalu-
ation of memory access performance in a multi-threaded system,
where M threads independently attempt to access a shared
memory segment. The model assumes that memory access times
follow a Poisson process with exponentially distributed inter-
access times. Key performance metrics, such as the probability of
no contention, the probability of K threads contending, and the
expected delay in accessing memory, are derived and analyzed.
The results show that contention significantly increases with the
number of threads, leading to sharp declines in performance.
Numerical evaluations highlight the exponential increase in
contention and memory access delays as the number of threads
grows, emphasizing the importance of efficient synchronization
mechanisms for scalable multi-threaded systems.

Keyphrases: memory access, multi-threading, performance evaluation, poison process, stochastic model

@booklet{EasyChair:15279,
  author    = {Brahmaiah Gandham and Salma Shaik and Rama Murthy Garimella and Praveen Alapati},
  title     = {Performance Evaluation of Multithreaded Systems},
  howpublished = {EasyChair Preprint 15279},
  year      = {EasyChair, 2024}}