Mechanisms for Adaptive Use of Mobile Network Resources

Wireless networks, mobility support optimisations

Computer communication networks are a very exciting field of research. Despite their widespread use, they are still in their infancy. Moreover, the recent use-cases have not been foreseen in the initial designs. A lot of clean-slate proposals can be proposed to address them but to be usable, an incremental path needs to be offered in order to provide compatibility with the other communications peers.

The goal of the research of Olivier Mehani was to investigate how to support user applications for current mobile terminals by selecting the best network configuration under competing requirements of best applications quality, lowest battery use and reasonnable access prices.

           
               Fig 1: Cross-layer control framework                    Fig 2: OODA loop for network adaptation
Various criteria

The thesis of Olivier Mehani has focused on proposing, on top of existing technologies (IPv6, mobility, comodity hardware), a solution able to take into account high-level requirements and expectations (good video quality,timely information), matching the relevant application parameters (codec, reporting frequency) to the characteristics of the available wireless networks (capacity, data transfer cost), and selecting the parameters of the entire network stack in order to optimise these various criteria.

Such a solution, it came to pass, requires addressing many orthogonal problems. How should the information (network characteristics, device performance, application quality profiles, user's expectations) be collected in generic enough a way that it can be effectively considered by some decision system? The question of which decision system to use, which information to consider and what to optimise was also a corner stone of the work. Finally, once some decision is made, the problem of applying this global decision to the multiple stand-alone components of the network stack, and adapt them to use-cases which they might not have been designed for in the first place, needs answers.

OODA loop

A model of a control system, the OODA (Observe, Orient, Decide, Act) loop was followed to create a modular cross-layer framework. The decision question was narrowed down and defined as the multihomed flow management problem (MFM): Which network(s) shall we connect to, and how should the application flows be distributed on the established links? A solution to this problem was proposed as the Quality-aware MFM, which selects the solution optimising user-perceived application quality (adapting application parameters as fits), preserving the battery lifetime and only incurring the most necessary monetary costs.

The action problem of network stack elements adaptation was addressed next, with a study of a transfer rate control protocol, usable with real-time traffic such as video streaming, in mobility situations where connectivity is intermitent and variable. Enhancements to support quick adaptation in these environments were then proposed. Finally, the problem of reporting was investigated, with the repurposing of an off-the-shelf software library, and the design of a thorough experimental protocol to ensure that it did not have any statistically significant observer effect in this context and was therefore fit for the purpose of online reporting in a tight (OODA) loop.

Working towards this contribution  also showed that, at some point, some of these underlying technologies need to be gracefully gotten rid of (e.g., IPv4), or other abstracted further downthe stack layers (e.g., socket API) to foster further evolution.


Fig 3: Application parameters need to be matched to network parameters (here: H.264)
Fig 4: Without proper adaptation, changes of network overly disrupt the quality of applications
Olivier Mehani's path

After an Engineering Technologist Degree in Electrical Engineering andComputing (DUT GEII), Olivier Mehani obtained an Engineering Degree from UTC. Then he got the opportunity to do a Masters Degree in parallel with his last year, in Sweden (Chalmers Tekniska Högskola). His Masters' thesis was done in one of Inria's research labs (Imara), dealing with Artificial Intelligence for autonomous vehicles, which confirmed his attraction to science and gave him a taste for research.

After his Masters'Thesis, Olivier Mehani worked for a year on such an infrastructure for Intelligent Transportation Systems, but wanted to dig deeper into the question, and have more scientific foundations to his work. So, the need to pursue further studies to be able to explore this at length became increasingly obvious for him. He got the opportunity to start a PhD in cotutelle with Robotics Centre (MINES ParisTech - CAOR)  and the University of New South Wales (Sydney, Australia).

This altogether gave him many opportunities for collaboration and exposure to a wide range of people, skills and approaches. He was able to apply his previous knowledge to more generic problems, and really got to understand what science and a researcher's work are about. Olivier is now a researcher with Nicta's Network Research Group, in Sydney, Australia and amongst other things, working on follow up work of his thesis.