CAREER: Rethinking Abstractions in Virtualized Architectures and Systems

University Of Texas At Arlington, Arlington TX

Investigators

Abstract

Recent advances in cloud computing have led to a wide adoption of virtualization techniques in modern computer systems which allow better utilization of computer resources. There is also a steady trend towards building future data centers and high-performance computers with a software-defined architecture. However, challenges remain in adopting virtualization in many critical domains. First, there still exists a large performance gap between virtualized and physical systems, especially for high-speed devices, scientific workloads and latency-sensitive applications. Second, performance isolation and resource elasticity are often contradictory goals under the current models, which leaves much of the economic benefit of virtualization unexploited. Third, the additional system complexity in virtualiation undermines stability and predictability. This research addresses these issues and seeks to improve the performance, cost-effectiveness, and predictability of virtualized systems. The research will be tightly integrated into teaching and further broaden its impacts through mentoring and recruiting minority students, and outreach activities in K12 schools. Specifically, this project will identify gaps in the existing abstractions that cause performance degradation, inefficiencies and unpredictability, as well as pinpointing the essence of current abstractions that has enabled isolation, modularity and portability. The project entails three research thrusts: First, it will design and implement augmented abstractions for various types of virtualized systems, including virtual machines, containers and virtualized networks, to bridge the semantic gaps. Second, it will leverage the augmented abstractions to design efficient, effective and elastic resource management schemes while retaining much of the benefit of the existing abstractions. Third, it will increase the understanding of abstraction in multi-tenant systems and apply the knowledge to studying the inefficiencies of conventional systems and guide the design of new abstractions in emerging architectures. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.