TSSG recently completed AMEND, an industry-targeted research project with Intel Research Lab, Leixlip under the auspices of the CONNECT Research Centre for Future Networks and Communications. The main contacts in Intel Labs were senior researchers John Kennedy and Dr Radhika Loomba, who studied for her PhD here in WIT/TSSG before taking up her position in Intel. The project was led in TSSG by Dr Brendan Jennings and Dr Bernard Butler, who jointly supervise the PhD students Ms Kanika Sharma and Mr Adnan Mahmood who were funded by AMEND.
AMEND’s objectives were to examine how, in software-defined infrastructures, (both networks and computing environments), resources (which include network bandwidth, computing and storage) can be managed so that services deliver their objectives.
The problems we addressed
The main challenges is that it is very difficult to manage both physical and virtualised resources together. This is because:
- the mapping between the two types of resources needs to change dynamically in response to external conditions.
- the architectural layers (physical infrastructure, virtual infrastructure, software) are deliberately isolated from each other, to simplify how they are designed and deployed, but managing them needs visibility across such boundaries.
Unfortunately, traditional resource management approaches are not agile enough to respond to such challenges. Therefore, we worked with Intel Labs to agree on a set of research scenarios where:
- virtualised networks are prevalent, or are expected to be so in the near future, and
- service quality needs to be assured and managed carefully
We identified Intelligent Transport Systems as a suitable domain for our study.
Theme 1: Vehicle clusters as a sensing platform in smart cities
Kanika is looking at the problem of using (opportunistic) clusters of vehicles as a sensing platform. Modern vehicles have an increasing range of onboard sensors, typically for driver assistance and/or entertainment. When vehicles travel together, it is possible to form once-off vehicle clusters. Within a cluster, vehicles can share their resources to collect data about their environment, to process it and then to forward it to internet gateways at the roadside. Such opportunistic clusters could be used to collect traffic patterns, pedestrian footfall, road condition, etc. Clearly, service components placed on vehicles in the cluster should respect the availability of resources, e.g., they must not saturate any wireless links! Furthermore, wireless connectivity within the cluster and with the roadside can be patchy, so the virtual networks need to change frequently to accommodate this. Lastly, cluster cohesion can be a problem, since vehicles join and leave the cluster at will, typically at road junctions.
We developed models for allocating service components to (vehicle) nodes in a way that maximises the likelihood that a service can be deployed successfully and is robust enough to deliver its sensing objectives. This work was published and we continue to expand and improve those models.
Theme 2: Reliable data transfer between fast-moving vehicles
Adnan is looking at the problem of ensuring that vehicles can share safety-critical data (both warning messages and video streams) even when traditional wireless connectivity is very poor. This is compounded when the vehicles are moving quickly (outside urban areas) and so have little time to exchange or forward this data. Interestingly, there are many wireless technologies but they have different strengths and weaknesses. For example, the automotive industry has proposed the DSRC (IEEE 802.11p) standard for short-range communications, but it has relatively low bandwidth. Cellular networks, when available, have greater bandwidth but their latency (delay setting up a connection) can be too long for safe communication between vehicles. Therefore horizontal handovers (between wireless networks of the same kind, say DSRC to DSRC) and vertical handovers (between wireless networks of different kinds, say cellular to DSRC) might be necessary to maintain adequate Quality of Service as vehicles move between coverage regions.
We developed a set of algorithms that decide when and how to handover to a different wireless network. We supplemented this with edge caching, which uses the extra storage and bandwidth resources on the roadside units (the network edge) to reduce the overall amount of data being transferred wirelessly between vehicles. This work was also published and continues to be developed.
In a world where 1) more and more contextual data is being collected in smart cities and 2) driver assistance systems and even self-driving cars are being promoted by the automotive industry, the management of the computing and networking resources is essential for success.
If vehicles can be used for environmental sensing, offering reliable data about a (smart) city while not harming the experience of the vehicles’ drivers and passengers, this has great potential benefits for society. We still need to solve the management issues, to make such services feasible, but we have identified the research questions that will help us get there.
Also, there is a societal need for road travel to be as safe and environmentally friendly as possible, so driver assistance, vehicle platooning and automation show great promise in this regard. However, there are many practical problems (e.g., ensuring that safety messages arrive in a timely fashion) to solve before the benefits outweigh the risks.