Content Delivery Networks (CDNs) are an important solution for easing Internet
network traffic congestion and improving response latency. There is a growing trend to deploy Cloud CDNs (CCDNs) to provide more flexibility compared to traditional CDNs. Today, regardless of the current technological advancements, most CCDNs still make their resource and cache management decisions while being unaware of what Content Providers (CPs) want to achieve. For instance, the APIs of prominent CCDNs (such as Amazon CloudFront, Google Cloud CDN, and Microsoft Azure CDN) do not allow CPs to express their high-level targets, e.g., requests/region. Instead, CPs are mainly limited to specifying the desired geographical coverage and the origin content server. Fortunately, there has been an increasing interest in the Intent-Based Networking (IBN) paradigm, which aims to allow users to express what they want to do (in the form of an intent) instead of how to do it. By employing IBN, CCDNs could move towards a communication scheme that is more adaptive, flexible, and portable,
that tries to automate meeting the CP’s intent target throughout the pre- and postdeployment phases of a service.

In this thesis, we present a solution to enable such Intent-Based CCDN. We
illustrate how microservices modularity could be utilized in the intent translation
process, decomposing intents into behavioral abstract policies that in turn get
translated into technical realizations as microservices. This provides a more flexible, cost-effective, interoperable solution, and enables stakeholders to compare and alternate between microservice alternatives depending on the CP’s intent. We propose and implement a Low-Cost Intent (LCI) that targets cost reduction while provisioning a CCDN.We evaluate several realizations of this LCI, namely LCI1, LCI2 and LCI3 via selecting different microservice alternatives (i.e., leading to different overall CCDN deployments) for each. Additionally, we propose and implement different LCI refinement algorithms for performance improvement in various traffic scenarios. We conduct our evaluation experiments on Google Kubernetes Engine (GKE), and test it using traces of real CDN traffic obtained from a major ISP. Evaluating our refined LCI different realizations against a baseline of GKE demonstrated the differences between these varying options. The best was LCI1 which reduced infrastructure cost by up to 10% in return for a 2% rise in dropped requests. On the other hand, although LCI2 also led to cost reduction up to 10%, it caused an increased request drops by almost 20%. Finally LCI3 led to the highest infrastructure cost reduction by up to 20% but in return for the highest rise in dropped requests by 40%. Interestingly, in a traffic bursts scenario, although the refined LCI resulted no cost reduction, it actually outperformed the baseline by reducing the amount of dropped requests by 30%. Accordingly, we discuss the factors that led to these different variations between the different LCI flavors based on their corresponding selected microservice alternatives.