Abstract

Cloud computing provides a wide range of virtualized services, such as computing power, storage, and applications, which can be accessed on-demand. Cloud-native applications are built using a microservices architecture, where independent, function-specific services communicate through APIs and messaging protocols. Service graph is a directed graph that shows interaction and dependencies between these services. An application can be achieved by using different services with same functionality which result in having multiple service graphs. However, selecting an optimal service graph among all possible ones is a challenging task that requires expertise in cloud applications, their dependencies, and compatibility, along with meet non-functional requirements such as bandwidth, and latency. Tradition
approaches rely on domain experts who manually generate service graphs, a time-consuming and error-prone approach
that often results in suboptimal configurations, higher deployment costs, and unmet performance criteria. These challenges highlight the need for automated solutions to efficiently produce optimal service graphs aligned with cloud consumers expectations.
Intent-Based Networking (IBN) is a new paradigm that allows cloud consumers to specify their requirements at a
high level, without dealing with the underlying technical complexities. Intent is a request at a high-level of abstraction (e.g., Natural Language), which describes what the cloud consumers expect. More specifically, it enables cloud consumers to focus on defining ”what” they need, such as performance targets, cost constraints, or latency requirements, without needing to specify ”how” these objectives should be achieved. Instead of manually selecting and configuring services, cloud consumers express their intent in terms of objectives such as performance targets, cost constraints, or latency requirements. Then, these high-level intents translates into low-level configurations, automatically generating and deploying the service graph in the cloud.
In this thesis, we address the problem of service graph generation and selection while considering both functional and non-functional requirements derived from cloud consumer intent. Our objective is to minimize the total deployment cost of the service graph in a data center network and to determine its placement within the distributed data center network. To address the problem, first, we translate high-level intent using a domain ontology into its functional and non-functional requirements which are specified in terms of initial services, latency, and bandwidth. Then, we use a service catalog along with the initial services to generate all possible service graphs that can meet the functional
requirements of the given intent. We formulate the problem as an Integer Linear Programming (ILP), taking into account the bandwidth and latency requirements of the intent. To solve this, we propose our Service Graph Selection (SGS) solution, which aims to achieve a near-optimal solution in a computationally efficient manner. Our results demonstrate that the proposed solution achieves a deployment cost that is only 4-6% larger than the lower bound of the optimal deployment cost.