Download files
Abstract
In past recent years, we have observed that the number of smart devices worn and carried by users is growing rapidly driven by innovative new smart wearables and interesting service offerings. The ubiquitous adoption of wearables is truly changing the way mobile computing is deployed and mobile applications are being developed. This has led to applications that utilise multiple devices around the body to provide immersive environments such as Mixed Reality. These applications rely on a number of different types of functions such as sensing, communication and various types of processing, that require considerable device resources such as energy, computation and memory. Also, many current applications require access to Internet data. However, these resources are much limited in the hand-held and wearable devices thus limit the applications from not achieving its full capability and leaves the users dissatisfied. Ability to incorporate more resources on the device is again limited as it likely to make the devices bulky and more expensive. Therefore, efficient utilisation of available resources is paramount.
Today, a person owns multiple of these smart hand-held and wearable devices that are interconnected and form a Personal Area Network (PAN). When the same function is available on multiple devices that form part of the PAN, and/or is in close proximity (i.e., common-functions), it is possible to consider them as a re- source pool that can be shared among the devices. This enables the applications to dynamically leverage the capabilities of the devices in the PAN to optimise resource usage, thereby elongate the battery uptime, provide the best quality of service, minimise the monetary cost of usage, depending on the user/device context and application requirements. To harness the collective capabilities of PAN devices at present, each application has to implement utilisation of the distributed resources, by explicitly considering the availability of the devices in the PAN and the supported functions, the cost of accessing each of the function and their quality of service, and the dynamic changes in the context of user and devices. This requires a wider understanding of distributed PAN systems at the initial phase of the application development, which is nearly impossible and increases the complexity of application development. Therefore, a platform takes user and device context into account, facilitates applications to utilise all PAN resources intelligently and automatically, and dynamically adapts to the context changes is needed. We introduce the concept of Application Function Virtualisation (AFV) that enables each application seamless accessibility to the available resources within the PAN and provides intelligence. AFV takes care of monitoring context changes, optimal decision making depending on the context changes, and the dynamic adaptation of the system.
The thesis first shows that the wearable devices and applications are being adopted by users and developers in the same way as smartphones. Also, users and the applications have even more mobile behaviour than in smartphones that are more crucial in day-to-day usage. Therefore, we can predict the wearable usage will increase significantly that the users will have multiple of the PAN devices. Next, we experimentally demonstrate that the current PANs do not utilise application functions and resources intelligently. Also, we present the current practices of the execution of the application functions in different devices and analyse how they can be efficiently executed at different devices. Then, the thesis presents the architectural design of Application Function Virtualisation (AFV) that enables seamless access to the pool of functions and automated dynamic function virtualisation/scheduling across devices. Also, AFV enables intelligent resources utilisation by simplifying the context-aware application development. AFV provides a simple set of APIs hiding complex architectural tasks and continuously monitors context/application requirements. We also demonstrate how the virtualised functions can be used to optimise the performance of a PAN based on the user/developer defined objectives. Finally, the feasibility of AFV design is shown by implementing AFV on Android and evaluated in terms of energy consumption, latency and communication over- head. Moreover, the benefits for the user in terms of resource efficiency and quality of experience with relevant use cases are evaluated.