Middleware Support for a Dynamic and Safe Internet of Things

The Internet of Things (IoT) bridges the physical world with the virtual world by embedding everyday objects with computation, communication and sensing abilities, interconnecting them with each other and the internet. As the IoT rapidly gains in popularity, the size and complexity of deployments is also increasing. Prime examples are smart cities and Industry 4.0, where the return on investment of large-scale infrastructure is improved by running multiple applications with ever changing requirements concurrently throughout the networks lifetime. Furthermore, with IoT technology becoming ubiquitous in our daily lives, security becomes a primary concern that is likely to grow in complexity as we move to more dynamic environments. Dynamism at large scale requires fundamental support for software evolution and adaptation from the underlying software stack. State-of-the-art programming abstractions and reconfiguration approaches partially accommodate this through modular software development and runtime remote reconfiguration of distributed applications. While the advantages are clear, these solutions fall short when managing complex decentralized configurations. Incompatible configuration causes faults and downtime, and remote inspection and reconfiguration of applications incurs a considerable message overhead and energy cost on low power wireless networks. Contemporary solutions to secure the IoT require in-depth hardware modifications, raising the cost of roll-out for the millions of devices already active in the field and preventing their adoption. This dissertation presents three contributions towards a dynamic and safe Internet of Things. The first contribution focuses on detecting and avoiding inconsistent configuration of distributed IoT applications. Safe reparametrization offers a descriptive language to developers for expressing distributed configuration dependencies, and uses a network protocol to resolve and enforce these dependencies at runtime. The overhead of this approach on representative hardware platforms is minimal, while management effort is significantly lowered when compared with exhaustive methods of ensuring safe and correct reconfiguration. The second contribution reduces the overhead and complexity of inspecting and reconfiguring distributed IoT applications through two novel approaches. Refraction selectively augments application data flows with meta-data, which travels to refractive pools without additional network traffic. Refractive pools serve as natural loci of low-cost inspection and control for distributed applications. Reactive policies deployed to these pools allow automatic reconfiguration based on changes in application state. Tomography further reduces overhead by recognizing that software components bound together are often queried in groups, using probes to traverse component compositions to collect meta-data and enact change efficiently. Prototype implementations have been evaluated within a smart office deployment, and show notable improvements in terms of management complexity and network traffic. The final contribution looks at cost-effective security for the IoT. The Security MicroVisor is a pure-software approach that provides memory isolation for low-end IoT devices through partial virtualization of the instruction set. Subsequently, this memory isolation is used to safeguard the integrity of critical secure operations and secret key material while concurrently running insecure user applications, effectively providing a Trusted Computing Base. Evaluation shows that for typical usage patterns of IoT applications, virtualization latencies are imperceivable, while resource overhead is small.status: publishe