Date of Award
Doctor of Philosophy (PhD)
Recent years have witnessed the emergence of edge computing as an enabling platform for time-sensitive services. However, existing edge computing platforms face a multitude of challenges in meeting the latency requirements of time-sensitive applications. (1) Traditional real-time virtualization platforms require offline configuration of the scheduling parameters of virtual machines (VMs) based on their worst-case workloads. However, this static approach results in pessimistic resource allocation when the workloads in the VMs change dynamically. (2) Edge computing operators must deliver consistent tail latency performance for time-sensitive applications deployed on different edge sites. Traditionally, significant effort is required to test, tune, and configure a time-sensitive service for each edge host. This is a labor-intensive process that cannot scale effectively for a large number of edge sites. (3) Many time-sensitive services need to handle aperiodic requests for stochastic arrival processes, which differs from periodic and sporadic models in traditional real-time systems. It is necessary to provide a new latency analysis for predicting the latency distributions of those aperiodic requests on an edge environment. (4) Fault-tolerant coordination for maintaining consistency in distributed applications. However, traditional failure recovery approaches employed by coordination services incur excessive recovery latency unacceptable to time-sensitive applications.
This dissertation makes the following contributions to the field of real-time edge computing through the design, implementation, and experimentation of four novel system technologies and architectures.
Multi-Mode-Xen (M2-Xen): Dynamic CPU Resource Provisioning. M2-Xen enables dynamic resource allocation on a real-time virtualization platform, allowing VMs to switch modes with different CPU resource requirements at run-time while maintaining desired real-time performance.
Virtualization-Agnostic Scheduling (VAS): Scheduling for Consistent Latency. VAS provides a novel scheduling framework to maintain similar latency distributions for time-sensitive tasks on different virtualized hosts.
Stochastic Response Time Analysis for Aperiodic Tasks on Virtualization Platforms. To achieve predictable latency for aperiodic time-sensitive services, we establish a novel queueing model, M/D(DS)/1, for stochastic response time analysis of aperiodic services following a Poisson arrival process on computing platforms that schedule time-critical services as deferrable servers.
RT-ZooKeeper: Fast Recovery in Replicated Coordination Services. RT-Zookeeper employs novel leader election and recovery protocols to address the limitations of Apache ZooKeeper, the prevailing open-source coordination service. Implemented based on ZooKeeper version 3.5.8, RT-ZooKeeper significantly reduces the latency in recovering from leader failures in replicated coordination services.
Christopher D. Gill, Sanjoy Baruah, Ning Zhang, Linh Phan,