A Subscriber-Oriented Interference Detection and Mitigation System for Cloud-Based Web Services
Date
2018-05-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Web services are increasingly being hosted on public cloud platforms. Such platforms often employ resource virtualization. Specifically, a public cloud provider can co-locate multiple virtual machines (VMs) belonging to different cloud subscribers on a single physical machine (PM). The responsiveness of a Web service hosted on a VM can be impacted by interference, i.e., contention among multiple VMs for shared PM resources. Response time degradations can frustrate end users of the Web service, who may eventually terminate the service. Most existing studies have focused on developing techniques that cloud providers can use to mitigate the adverse impact of interference. In contrast, very few studies have focused on directly enabling cloud subscribers, i.e., Web service operators, to address this problem. Developing subscriber oriented techniques is challenging since subscribers typically do not have direct access to PMs and hence cannot easily measure inter-VM contention. This thesis focuses on addressing this challenge. Specifically, it develops a system that can dynamically adapt a Web service at runtime so that it satisfies a specified response time target even in the face of fluctuations in service workload and interference. Using a benchmarking study on a popular public cloud platform, I show that the impact of interference can be significant onWeb service response times. I then develop a novel software probe based system that helps a cloud subscriber obtain a direct measure of how interference is currently impacting service response times. I also develop a runtime management controller that exploits the probe to automatically scale in and scale out resources to handle changes in workload and interference. Results from both public and private cloud setups indicate that the probe technique is very effective in detecting the extent and type of interference. Furthermore, the controller ensures that a desired response time target is met at each resource instance in a system while using only the minimum number of instances necessary given the observed workload and interference conditions.
Description
Keywords
Citation
Mukherjee, J. (2018). A Subscriber-Oriented Interference Detection and Mitigation System for Cloud-Based Web Services (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/31905