Design of Scalable Multi-core System Strategies

Abstract

In pursuit of ever growing performance, more and more
processor architectures have turn out to be multicore processors
system. As clock frequency was no longer escalating rapidly and
ILP techniques showed retreating results, growing the amount of
cores per chip was the usual choice. The transistor resources is still
rising and thus it is predictable that within ten years chips can have
hundreds of high performance cores. Scaling the amount of cores,
however, does not essentially convert into an equivalent scaling of
performance.
In this paper, we recommend numerous stratgies to get better the
performance scalability of multicore systems. With those stratgies
we address a number of challenges of the multicore area. First, we
explore the result of the power wall on prospect multicore
architecture. Our representation includes predictions of knowledge
improvements, investigation of symmetric and asymmetric
multicores, as well as the authority of Amdahl’s Law. Second, we
investigate the parallelization of cores application, thereby
addressing application scalability. Presented parallelization
strategies are discussed and a novel strategy is planned. Analysis
shows that using the new parallelization strategy the number of
available parallelism is in the order of thousands. A number of
implementations of the strategy are discussed, which show the
complexity and the prospect of actually exploiting the available
parallelism.