Model-driven adaptation of Function-as-a-Service applications

Abstract

This doctoral dissertation is situated in the research field of FaaS processing, cloud topology adaptation and administration. It consists of an approach suggesting the definition of a cloud and edge processing topology using the TOSCA standard. Also, the presented approach includes a novel way of updating the processing topology based on the criteria set by the DevOps. Concerning the description of the applications, research on the most important works from the state of the art was carried out, focusing on contemporary generic application description languages. Also, the expressivity of these application description languages is briefly presented, and a comparison of the modelling extensions and approach suggested in this work was made against one of the most important and ubiquitous description languages (Terraform). Related to the adaptation of applications, research on the available adaptation methods was performed, and a new rule-based methodology suggesting the use of ‘Severity’ of the situation of a topology is described. Severity values are obtained by factoring in all of the dynamic metric values involved in a violated rule, and based on these a relevant adaptation action is suggested. The exploitation of Severity values using different algorithms (techniques) allows different ways of countering workloads. In the context of the aforementioned research four software systems were created, of which two are open-sourced. The first is related to the creation of application topology descriptions, leveraging the new TOSCA extensions. The remaining are related to analysis of Severity techniques and the creation of adaptation actions based on Severity techniques, although in two of them additional techniques which are widely used in the industry were also implemented. One of these systems is a simulator. Severity was proven to allow the definition of techniques leading to improved results, as these were determined by a chosen utility function. The experiments which were carried both in the level of simulations but also in a realistic testbed, indicate the need to make appropriate decisions on the technique which should be used based on the workload which is served. The successful use of Severity in the context of OpenFaas, a well-known and realistic FaaS platform, to perform the adaptation indicates the feasibility of the approach.