Doctor's Theses (authored and supervised):

C. Avasalcai:
"Quality of Service aware Resource Management for Edge Systems";
Supervisor, Reviewer: S. Dustdar, F. Leymann, G. Pallis; Institute of Information Systems Engineering, Distributed Systems Group, 2021; oral examination: 2021-09-24.

English abstract:
Demanding latency-sensitive applications have stringent requirements such as high availability and low latency. The current cloud-centric systems face challenges in satisfying the applicationīs stringent requirements. As a result, researchers have proposed two new paradigms, i.e., edge and fog computing, as an alternative to deploying demanding IoT applications closer to the edge of the network. By extending the cloud system with these two paradigms, we obtain an edge system. Edge systems have been identified as a solution to distribute more resources closer to the end-user since meeting application demands must occur at runtime, facing uncertainty, and in a decentralized manner. However, the edge systemsī distributed nature makes the application development more challenging since the developer must divide the applicationīs functionality into multiple
microservices. Furthermore, high volatility defines the edge system due to edge nodes being characterized by (i) heterogeneity and (ii) mobility, making a node unreliable - a node may fail or leave the network unexpectedly. As a result, the application deployment
and management under volatility is more challenging. This calls for novel application development methodologies and resource management techniques that comply with the applicationīs requirements and aids the developer to develop, deploy, and manage an application in the target edge system. However, developing these techniques is not a
trivial task.
In this thesis, we provide novel methodologies and resource management frameworks to enable the efficient utilization of the edge node available resources and maintain the correct application functionality throughout its entire execution. Our objective is to (i) aid the developer during the application development process, (ii) deploy the
latency-sensitive applications in the target edge system, and (iii) ensure that deployed applications remain operational during their lifespan. We start by proposing EdgeFlow, a new methodology for latency-sensitive IoT applications development and deployment
on the edge system. The purpose of EdgeFlow is to assist the developer during the application development process by allowing the developer to define the application requirements and validate them at design time. Three different stages characterize our proposed framework, i.e., the (i) development, (ii) validation, and (iii) deployment. To this end, we propose an extension of the Flow-Based Programming paradigm with new timing and resource requirements. Moreover, we provide a resource management technique to assist with the deployment and validation stages. In the next part of our thesis, we focus
on providing a novel decentralized resource management technique and accompanying technical framework to deploy applications on resource-constrained devices. The proposed framework enables the efficient utilization of available resources distributed between resource-constrained edge nodes. The resource management framework uses satisfiability to find at runtime a deployment for an application; the mapping produced is compliant with microservices resource requirements and the application latency constraints by construction. Our approach ensures seamless deployment at runtime, assuming no designtime
knowledge of device resources. We further propose a new robust IoT application model that follows a hierarchical architecture. We model an application component using multiple configurations - each configuration has a different functionality level and resource requirements. Additionally, we extend the decentralized resource framework to be able to deploy the new application model. Finally, we propose an adaptive framework capable of efficiently deploying and maintaining a microservice application on an edge system. Our framework tackles two intertwined problems - (i) finding a microservice placement across devices and (ii) building an invocation path that serves the deployed application. For this framework, our objective is to maintain the correct applicationīs functionality by satisfying its given requirements in terms of end-to-end latency and

Decentralization / Edge Computing / Resource Management / Application development / Internet of Things / Self-Adaptive

"Official" electronic version of the publication (accessed through its Digital Object Identifier - DOI)

Created from the Publication Database of the Vienna University of Technology.