phd Christophe Huygens

Friday March 2, 2012 -- 16h30 CET
Public PhD defence Christophe Huygens -- promotor: prof. dr. ir. Wouter Joosen

Celestijnenlaan 200A
Ground floor auditorium 00.225

Supporting heterogeneity of stakeholders in low-resource networked systems.

Abstract

Wireless sensor and actuator networks (WSANs) are seeing increased acceptance in the real world because of their potential to increase productivity. These wireless but limited systems can be applied in a wide range of problems ranging from logistics to e-health, due to their moderate cost which promotes redundancy and close observation of phenomena as well as network-level self-organisation which promotes mobility and eliminates the need for infrastructural engineering. Yet, this class of systems is not without its challenges: energy as well as resources are limited, the individual node is unreliable and the total number of nodes is high. Mentioned limitations are traditionally mitigated by careful engineering of the single application using the WSAN. Existing research has focused on providing optimal support for this monolithic usage mode. Real-world application however are generally not monolithic - driven by cost-effectiveness, a single sensor (network) must support multiple applications of diverse users and nodes of networks owned by different parties must work together to achieve users' goals. The WSAN ecosystem is shared since users are competing for the scarce resources, as well as federated since finishing a task will often require cooperation spanning logical network boundaries. Improving support for shared and federated operation of WSANs is the overall goal of this dissertation.

The envisioned federated setting with multiple owners and users of networks raises additional requirements that are mainly security-centric from a runtime perspective. Fine-grained control of resource consumption is needed, next to provisions for modularity and dynamic reconfiguration that were already established through multi-application but single-owner usage scenarios. Trust established between the owners needs to be translated into the federated network so that services and applications on nodes can cooperate. The runtime perspective however only offers a partial view to the federated problem. The multiple applications of users are competing, but also asynchronously coexisting. Applications come and go at a different pace. This is not different from traditional multi-user systems where this problem is tackled by managing the lifecycle of each application independently culminating in resource-expensive virtualisation - now basically replicating a full host OS for each application. In the low-resource WSAN, communality must be embraced and abandoning these lifecycle silos implies that the traditional individual sequences of develop > deploy > manage cannot be maintained. Instead a system-software continuum is proposed wherein all stakeholders submit application or system goals. Optimally translating (from a resource perspective) these goals to low-level artifacts and the enactment of the latter is a prime responsibility of "enhanced" WSAN middleware. This middleware enhancement is also a suitable place for autonomicity to address the impact of mobility and node failure.

The contributions of this dissertation can be summarised as follows. The federated ecosystem is characterised and its requirements identified. To address these requirements from a runtime perspective, a middleware is proposed combining a component model and a policy-driven paradigm, with specific extensions for secure cooperation. This runtime middleware provides a modular, reconfigurable, and shareable execution environment for our target platform. Next, focus moves to the development support that is needed in the middleware from the continuum-lifecycle perspective. Suitable abstractions are proposed for the stakeholders of the federated ecosystem and the gap between those abstractions and the runtime execution environment is explored and addressed through a high-level description of the enhanced middleware. Finally, a subset of the enhanced middleware is presented in detail and validated. The latter demonstrates the optimisation potential of the continuum approach.

The language of the defence will be English.

If you intend to attend defence and/or reception please register here.

Getting there.

Printed copies of the text will be available on request only -- the text will be here for download.

Informally, I believe Moore's law has tricked us into a situation where we don't really care about resources anymore for most practical computing purposes. Because resources are plentiful and available, and most problems can be contained by isolating the applications from each other, we just give each application its own environment and throw resources at the problem. We have been doing this increasingly over the last 20 years and have become very good at it - so we never complain about too much memory or CPU. When confronted with a situation where resources are scarce, like in sensors where every byte or joule counts, our focus needs to shift from isolation to reuse - with all associated risks and inconveniences. I'd like to present you my thoughts on how I believe we can build and operate these systems, whilst still keeping the mentioned issues under control.

Security is one of the key benefits of isolation. Reorganising the system with less isolation means introducing security controls at other places. To achieve resource control or controlled interaction, we propose to use mini-firewalls acting on the data being exchanged between or on the sensor nodes. Walled garden activities in the resource-rich tier will be useful as well. In a real setting, these sensor nodes may belong to different members of a federation, so merging the nodes into a unified infrastructure needs to be tackled as well. These are all functions of the runtime middleware.

Even with the rich set of functions offered by the runtime middleware, a significant gap remains between the platform functions and the high-level, business-centric goals of users or owners. Here, we see an enhanced role for middleware - the middleware must collect all these goals and translate to a configuration (or allocation of modules to the nodes) that can be executed by the runtime middleware. Reuse, optimisation and self-management can also be accomodated in this enhanced middleware.