Thread Scheduling In Multiprocessor Systems Research

With the emergence of hyper-threading and multi-core processors, current SMP system hardware structure becomes much more complex and more powerful, the ensuing question is: traditional SMP operating systems no longer give full play to the performance of modern SMP hardware, in order to adapt to this change, we need to study traditional SMP system and improve those parts that are no longer suitable with new algorithms.We hope software could gain maximum performance by these study.First, this paper first introduces the research background and research significance, gives brief description of two main techniques to achieve parallel execution of multiple processes or threads in a single processor package: simultaneous multi-threading technology (SMT) and multi-core technology (CMP), in additional, paper also outlines the linux operating system and its important role in the study of smp system. Second, paper analyzes the organizational structure of classic smp system and introduces a variety of process scheduling algorithms; knowledge of these hardware architecture and software concepts gives us solid foundation in further study of smp operating system. Third, paper uses the latest version of linux kernel as an object of study and gives detailed analysis of its process scheduling system. The current linux kernel uses an entirely new scheduler called completely fair scheduler (CFS), in order to understand its principles and related support for smp systems, paper gives detailed analysis of CFS's source code. Finally, paper summarizes several key factors in the design of smp scheduling system and presents an efficient scheduling model.Based on theoretical analysis in previous chapters, paper also presents an optimization proposal for smp system and set an example for how to make optimization to smp system and how to realize your optimization proposal.