User-centric cloud application management

Abstract

The accelerated growth in cloud computing technologies in the past decade has empowered a wide spectrum of cloud services, and this has originated a big challenge for cloud users, specifically for Infrastructure-as-a-Service (IaaS) users who find it difficult to choose a particular Virtual Machine (VM) type from the Cloud Providers (CP). Cloud application performance variability and IaaS pricing diversity amongst the CPs are seen as the main reasons for cloud users’ perplexity with selection of a VM type. Although cloud researchers have proposed various effective VM selection algorithms to support cloud users, these algorithms consider only the instantaneous performance of the cloud applications on the VMs and ignore whether the performance or Quality of Experience (QoE) will be maintained by the VMs in the future. That means they only consider the pre-deployment phase of the applications’ lifetime and ignore the post-deployment phase. It has been reported by researchers that cloud applications show performance variability in the post-deployment phase, which may result in QoE degradation. To maintain QoE cloud users may need to migrate their applications to a new VM type with higher configuration from the same CP or with a similar configuration from a different CP. Apart from performance, cost can also be an important factor for certain budget-constrained users who may be interested to migrate to different instances if the price for the current cloud service rises or other providers offer a better price. To help cloud users in detecting QoE degradation and selecting a target VM type for application migration, this thesis proposes a user-centric cloud application management framework, named MyMinder (Multi-objective dYnamic MIgratioN Decision makER). Specifically, MyMinder employs a ‘monitoring system’ to monitor application domain specific performance metrics, a novel ‘detection algorithm’ to detect QoE degradation and a ‘VM selection method’ to choose a target VM type for application migration in the post-deployment phase. Evaluation of the performance of the detection algorithm was carried out by deploying a Darwin media streaming service (representative of real-world media streaming applications), and by designing a benchmark for generating realistic media streaming client requests. Evaluation of the VM selection method was carried out by using an example scenario where a cloud application suffers from a QoE degradation and requires migration. The scenario was investigated in a lab-based cloud testbed where data analytic and HPC applications were executed. To represent real-world cloud service offerings from multiple public CPs, the testbed contains a number of server machines offering a range of VM configuration choices. The evaluation results demonstrate that the proposed detection algorithm can reliably detect the QoE degradation of a cloud application and the VM selection method can choose an optimal VM to migrate the application from the under-performing VM in order to maintain the QoE.

Date of Award	Dec 2020
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Engineering & Physical Sciences Research Council & EC-Horizon 2020
Supervisor	Ivor Spence (Supervisor), Peter Kilpatrick (Supervisor) & Dimitrios Nikolopoulos (Supervisor)