DYNAMIC, REAL-TIME VIRTUALIZATION AND CLOUD COMPUTING
Abstract
Recent years have witnessed two major trends in the development of complex real-time systems. First, to reduce cost and weight and enhance flexibility, multiple systems are sharing common computing platforms instead of being deployed separately on physically isolated hosts. Second, instead of being developed from scratch, complex real-time systems increasingly are being created by integrating other independently developed systems. However, such integration of systems atop common computing platforms raises significant new research challenges in simultaneously meeting the real-time performance requirements of multiple systems sharing common platforms, and in coordinating real-time performance requirements across increasingly distributed and dynamic systems. Our initial work on real-time virtualization platforms and compositional scheduling techniques has established key foundations towards meeting those challenges, and has followed a rapid transition from research results to publications in top-tier research conferences and journals and to open-source release within widely used virtualization environments such as Xen. Even so, many crucial research problems still remain open, and the work we propose here will address those further challenges through a new integrated research plan that combines (1) new advances in compositional real-time scheduling theory to deal with cache effects and other overheads encountered by real-time virtualization platforms in practice, (2) new multi-mode, dynamic, and adaptive resource management capabilities for real-time virtualization platforms, and (3) new infrastructure for managing real-time computing clouds based on those new dynamic real-time virtualization platforms. Specifically, this proposal aims to extend our previous research results with four new integrated research thrusts that will substantially expand the scope of real-time virtualization research and create new capabilities in real-time virtualization and cloud system technologies: (1) cacheaware compositional scheduling theory for real-time virtualization that incorporates realistic cache and system overheads on modern multicore platforms; (2) design of multi-mode virtual machines that can address mixed criticality levels; (3) a real-time cloud management architecture based on real-time virtualization and scheduling, including support for real-time-aware migration of virtual machines; and (4) implementation, experimenal evaluation, open-source release, and integration with widely used virtualization environments such as Xen. As was the case in our prior work, the proposed research is characterized by the tight integration of theoretical and systems research. The proposed systems research activities will provide a deployment platform within which to implement and evaluate experiments based on compositional schedulability analysis, an active area of research in real-time systems in which we have contributed practical techniques and platforms, giving researchers and practitioners new means for practical evaluation of their research results and a clear technology transfer path. Similarly, the proposed theoretical research offers opportunities to strengthen the capabilities of the real-time virtualization platforms by dealing with realistic overheads and establishing expectations for system behavior against which empirical results can be compared. Through these kinds of integration, the advances achieved in either the theory or systems areas of this proposed research thus will be mutually reinforcing.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 26, 2018
- Source ID
- N000141612195
Entities
People
- Insup Lee
Organizations
- Office of Naval Research
- United States Navy
- University of Pennsylvania