Inverting gravity: make the data stay put, moving computation to it, live migrating across the Continuum

Important changes are occurring in the digital world. One is infrastructural, as the devices in which computations may happen, the places where software can run, are increasing in number and power. Computers are now in vehicles, home automation units, and industrial systems, bringing the computation to the closest to the physical world and the data produced there. Another is functional and is due to the emergence of AI solutions that increase the ability of software systems and generate novel functionalities, such as autonomous navigation in vehicles. In short: we can compute more and in more places. There is great potential in this transformation, as moving computation closer to data sources improves latency and gives access to massive amounts of data. However, gaining benefits from these changes requires the ability to access and manage resources in a continuum-shaped infrastructure made of heterogeneous devices. Moreover, running software in the physical world forces applications to meet its requirements, which may be temporal, as for controlling a robotic manipulator, or related to regulatory constraints, as for the exploitation of data. Preserving the data proximity in a world made of moving objects requires a live migration mechanism capable of moving the computation where it is convenient to run. Furthermore, coordinating these migrations requires novel orchestration strategies capable of reacting, or acting in advance, of events in the physical world. These challenges are still largely unmet. This thesis intends to address them by proposing a novel computing model, known as the Compute Continuum.