Placement Strategies in Cloud Continuum Environments

The increasing deployment of Internet of Things (IoT) applications across diverse domains, such as healthcare, manufacturing, and smart cities, introduces new challenges in the efficient utilization of computational resources. To ensure timely execution of applications and reduce response times, effective application placement within the cloud continuum, comprising cloud, fog, and edge infrastructures, plays a critical role. The inherent heterogeneity of these infrastructures, coupled with the varying computational demands and data exchange patterns of IoT applications, makes placement decisions particularly complex. This thesis work investigates the application placement problem with a specific focus on heuristic-based policies designed to minimize the network delay experienced by applications, while efficiently distributing application modules across the available resources. Unlike optimization-based approaches, heuristics are lightweight, making them well-suited when fast placement decisions need to be taken. The proposed policies prioritize placing application modules closer to the IoT devices that generate and consume data, considering the computational capacity of available resources and the communication costs among inter-dependent modules. A key contribution of this work is the introduction of concept of resource overbooking, a mechanism that allows application modules to be allocated to devices even if their processing demands exceed the available computational capacity. This strategy leverages the tendency of devices to be underutilized. In fact, by employing overbooking, the usage of resources can be enhanced, possibly reducing reliance on cloud infrastructure and minimizing overall network delay. To address the multifaceted nature of the placement problem, two complementary heuristic policies have been devised. The Load-aware policy adopts a horizontal placement strategy across applications, giving equal priority to all applications. In contrast, the Resource-aware policy operates vertically, focusing on individual applications and leveraging overbooking. The effectiveness of the proposed policies has been thoroughly evaluated through extensive simulation experiments using an extended version of the iFogSim simulation toolkit and considering workloads with different communication and computational characteristics. Results showcase that the proposed placement strategies significantly reduce network latency compared to baseline approaches, while ensuring a balanced distribution of workloads across the available computing resources in the cloud continuum.