A Concurrency Modeling Technique for Performance Evaluation of Multiprocessor and Multicore Systems

Abstract

The analytical techniques are either highly inaccurate or often require unaffordable computational costs to evaluate the concurrency in complex multiprocessor and multicore systems. Recent studies show that simulation techniques are being used in engineering and scientific researches more than ever. Simulation technique can be a promising alternative to analytical techniques as the real-world complex and expensive systems can be modeled as simplified representations including only relevant aspects of the problem. As billions of transistors are available on a single chip, the total number of CPUs in a multiprocessor system and the total number of cores in a multicore architecture are expected to grow significantly. As the number of processing cores increases, currently available simulators will become inadequate to simulate the concurrency of future computation extensive systems. Therefore, we propose a simulation technique to effectively model the concurrency in multiprocessor and multicore systems. Proposed VisualSim accelerated concurrency modeling approach is able to evaluate system performance. We model concurrency in a multi-processor shared-memory system and measure two performance metrics, average response time and task completion time. We find the proposed concurrency modeling technique easy, flexible, and capable of determining deadlock and starvation, balancing loads, and scheduling tasks.