Electronic Thesis/Dissertation
 

Multi-Core Composite: Toward a Scalable and Predictable OS

Open Access

The increasing prevalence of multi-core and many-core architectures, together with thestagnation of single-core performance has motivated the investigation of parallelism andscalability. With multi-core architectures widely used in both server-end and embeddedsystems, requirements of processing high volume of data while maintaining low latencyemerge. As the computational infrastructure, operating systems provide system level sup-port to harness parallelism, and impact the ability of applications to efficiently and scalablyutilize the increased hardware capability and functionality.This thesis presents Multi-Core COMPOSITE (MC2), a component-based operating sys-tem which aims to provide both scalability and predictability on multi-core architectures.Three parts of MC2 are investigated in the thesis: 1) FJOS, an OPENMP implementa-tion which provides efficient parallelism and real-time support to applications; 2) SPECK,the u-kernel of MC2 which provides scalable and predictable kernel level services such asIPC and memory mapping / unmapping; and 3) PARSEC, an abstraction for scalable andgeneral data-structure access, synchronization and memory reclamation, which facilitiesscalable memory management in MC2 .Evaluations demonstrate that MC2 performs fundamental system management capabil-ities such as communication, thread management, and memory operations with both lowaverage-case overheads and bounded worst-case overheads, even at high core counts. TheMC2 kernel scales perfectly, and is the first OS kernel we know of that is entirely lock-less,and provides worst-case scalable TLB coherence. Comparing to traditional implementa-tions, MC2 has less overhead, is more scalable, and can make better use of parallelism.The multi-core Composite OS provides a strong foundation for scalable and predictablecomputation.

Author Language Keyword Date created Type of Work License
  • All rights reserved
Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL

Notice to Authors

If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.

Thumbnail Title Date Uploaded Visibility Actions
Preview of Wang_gwu_0075A_12759.pdf Wang_gwu_0075A_12759.pdf 2018-01-16 Open Access