Scalable Computing and Communications: Theory and Practice

9

–––––––––––––––––––––––

Modeling and Algorithms for Scalable and Energy-Efficient Execution on Multicore Systems

Dong Li, Dimitrios S. Nikolopoulos, and Kirk W. Cameron

9.1   INTRODUCTION

Traditional single microprocessor designs have exerted great effort to increase processor frequency and exploit instruction-level parallelism (ILP) to improve performance. However, they have arrived at a bottleneck wherein doubling the number of transistors in a serial CPU results in only a modest increase in performance with a significant increase in energy. This bottleneck has motivated us into the multicore era. With multicore, we can achieve higher throughput with acceptable power, although the core-level frequency of a multicore processor may be lower than that of a serial CPU. The switch to multicore processors implies widespread in-depth changes to software design and implementation. The popularity of multicore architectures calls for new design considerations for high-performance applications, parallel programming models, operating systems, compiler designs, and so on.

In this chapter, we discuss two topics that are important for high-performance execution on multicore platforms: scalability and power awareness. They are two of the major challenges on multicores [1]. Scalable execution requires efficient resource management. Resource management includes determining the configuration of applications and systems at a given concurrency level, for example, how many tasks should be placed ...