Serverless computing in cloud-edge scenarios

This thesis investigates extensions to serverless computing, and specifically Function-as-a-Service (FaaS), by broadening its scope beyond purely public, cloud-based architectures, towards edge systems and hybrid cloud-edge deployments. The inclusion of such environments brings several benefits, such as the reduction of data transmission needs and the creation of a cloud-edge continuum across which workloads can be transparently moved. Supporting these environments also creates the possibility to integrate Internet of Things (IoT) devices in a secure, transparent way. Incorporating edge systems, however, brings new challenges stemming from heterogeneous network topologies and resource-constrained devices. To address these, this work introduces several novel scheduling approaches, expressed through extensions of APP (Application Priority Policies), a configuration language for FaaS schedulers. These extensions, i.e., topology-aware APP (tAPP), affinity-aware APP (aAPP), and cost-aware APP (cAPP), respectively improve schedulers by focusing on the geographical distribution of workers, the affinity between functions, and the minimisation of the expected latency when communicating with external services. To the same end, the thesis also presents FunLess, an open-source serverless platform geared towards edge and cloud-edge deployments. The platform makes use of WebAssembly for function execution, leveraging its cross-architecture nature and the small size of its binaries to better target resource-constrained devices. Finally, the work explores an application scenario where drones act as mobile computational nodes within the platform. This system demonstrates real-time task execution and offloading in unreliable and mobile network conditions, validated through a simulated emergency response use case.