Author's School

School of Engineering & Applied Science

Author's Department/Program

Computer Science and Engineering

Language

English (en)

Date of Award

Summer 9-1-2014

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Chair and Committee

Chenyang Lu

Abstract

In recent years, we have observed three major trends in the development of complex real-time embedded systems. First, to reduce cost and enhance flexibility, multiple systems are sharing common computing platforms via virtualization technology, instead of being deployed separately on physically isolated hosts. Second, multi-core processors are increasingly being used in real-time systems. Third, developers are exploring the possibilities of deploying real-time applications as virtual machines in a public cloud. The integration of real-time systems as virtual machines (VMs) atop common multi-core platforms in a public cloud raises significant new research challenges in meeting the real-time latency requirements of applications.

In order to address the challenges of running real-time VMs in the cloud, we first present RT-Xen, a novel real-time scheduling framework within the popular Xen hypervisor. We start with single-core scheduling in RT-Xen, and present the first work that empirically studies and compares different real-time scheduling schemes on a same platform. We then introduce RT-Xen 2.0, which focuses on multi-core scheduling and spanning multiple design spaces, including priority schemes, server schemes, and scheduling policies. Experimental results demonstrate that when combined with compositional scheduling theory, RT-Xen can deliver real-time performance to an application running in a VM, while the default credit scheduler cannot. After that, we present RT-OpenStack, a cloud management system designed to support co-hosting real-time and non-real-time VMs in a cloud. RT-OpenStack studies the problem of running real-time VMs together with non-real-time VMs in a public cloud. Leveraging the resource interface and real-time scheduling provided by RT-Xen, RT-OpenStack provides real-time performance guarantees to real-time VMs, while achieving high resource utilization by allowing non-real-time VMs to share the remaining CPU resources through a novel VM-to-host mapping scheme. Finally, we present RTCA, a real-time communication architecture for VMs sharing a same host, which maintains low latency for high priority inter-domain communication (IDC) traffic in the face of low priority IDC traffic.

Comments

Permanent URL: http://dx.doi.org/10.7936/K7VM49BX

Download

Share

COinS