Access Providers Research Issues

Dynamic and Automatic Configuration of Collocated APs

The Personal Router project envisions wireless environments with a variety of network services offered by a diverse set of providers competing in the same physical spaces. While competing, providers have to cooperate to coexist efficiently. Even in environments without competition between providers, the high density of APs in some areas requires coordinated configurations.

This research therefore considers the problem of automatically configuring networks of 802.11 access points (APs) that share common physical spaces. The challenge is to find channel assignments for collocated APs that maximize the global network capacity and satisfy local service demands. The solution must satisfy the capacity constraints of the wireless spectrum and quickly converge to a stable and robust configuration under dynamic network conditions and varying service loads.

To address this problem we have developed an approach to reliably and effectively configure a collection of access points to provide the maximum possible level of service to users, to allocate capacity among different access points according to user demand, and to quickly respond to demand pattern changes caused by mobile or transient users. Our solution considers both inter-channel allocation to minimize interference between access points, and intra-channel allocation to minimize the competition for a shared channel by different access points when complete isolation is not possible. We address problems such as hidden terminals and interference "ripple effects" by building multi-hop configuration dependency trees and evaluating a "disruption cost" function to choose the most effective configuration as required.

Our algorithm is implemented efficiently as a set of distributed protocols. Each AP participates and locally decides an appropriate channel and estimated load level. An AP maximizes its share of the network capacity but cooperatively observes the algorithmically determined global priorities. By iteratively communicating these decisions, and the disruption cost of making changes, the global network of APs quickly stabilize to an efficient set of channel and capacity assignments.

Absent global coordination mechanisms, networks of APs today most often are statically configured by administrators who carefully consider channel conflicts and predicted service loads. Such static, offline configuration methods cannot react automatically to changing network conditions and place large management burdens on administrators. A few commercial 802.11 products attempt to automatically and dynamically configure themselves by considering the channel assignment of APs within their direct communication range. This single, restrictive configuration criterion ignores user demand, hidden terminals and often results in sub-optimal channel assignments.

We are evaluating our own approach along three fronts, formal analysis of the correctness of our protocols, simulation of large networks of collocated APs, and an implementation. We have proved the correctness of our protocol under reliable message delivery and proved that for all execution sequences our protocol converges to a correct solution. Our simulation results demonstrate that the protocol achieves the desired performance objectives. We are in the process of implementing our approach using Linux and the widely available HOSTAP software, which will allow us to evaluate its performance under real-world conditions.

To learn more read: QoS Assurance with Collocated Wireless Access Points

Designing Traffic Profiles for Bursty Internet Traffic

Access Providers offering wireless service to Personal Routers need to be able to provide services well suited to users actual needs. This research proposes a new class of traffic profiles that is better suited for metering bursty Internet traffic streams than the traditional token bucket profile.

A good traffic profile should satisfy two criteria: first, it should consider packets from a conforming traffic stream as in-profile with high probability to ensure a strong QoS guarantee; second, it should limit the network resources consumed by a non-conforming traffic stream to no more than that consumed by a conforming stream. We model a bursty Internet traffic stream as an ON/OFF stream, where both the ON-period and the OFF-period have a heavy-tailed distribution.

Our study shows that the heavy-tailed distribution leads to an excessive randomness in the long-term session rate distribu-tion. Therefore, it is inherently difficult for any profile that limits the long-term average session rate to give a strong QoS guarantee for the conforming traffic streams. Our sim- ulation demonstrates that a token bucket profile that couples the average rate control and the burst size control has a weak QoS guarantee. Based on this result, we propose a new class of traffic profiles that decouples the long term average rate control from the burst size control. Compared to a token bucket profile, this profile improves the level of QoS for a conforming traffic stream, yet limits the effective bandwidth" consumed by a non-conforming traffic stream.

Residential Resale of Wireline Broadband via Wireless

This research examines when the Personal Router vision would be economically in the interests of broadband providers. The proliferation of 802.11 and other wireless LAN technologies that take advantage of unlicensed spectrum offer a novel platform for extending broadband access and wireless services ubiquitously. This research explores the circumstances under which a wireline provider of broadband services (e.g., Digital Subscriber Line or cable modem) would find it advantageous to provide broadband services to a new customer via residential resale of wireline broadband services.

We examines business, policy, and industry implications of residential subscribers setting up WiFi base stations to share their wireline broadband connections with neighbors. This analysis assumes that appropriate mechanisms will exist to allow the residential resellers to charge for the services they provide, but that the upstream wireline carriers will not be able to separately monitor end-user and resold traffic. This research presents a cost model to evaluate the circumstances under which this scenario may be profitable for the wireline carrier.

We have identified two policy issues that could challenge our model of residential broadband resale. First, the availability of unlicensed spectrum. After comparing the regulation of unlicensed spectrum in Europe and in the U.S., we concluded that the U.S. offers a more attractive opportunity than Europe for experimenting with the residential broadband resale, mostly because the lack of transnational regulatory authority in Europe which has contributed to slow adaptation of national regulations and has impeded the development of wireless LAN products in Europe. Second, because spectrum is a scarce shared resource, its availability for each user depends on the number of other users using it concurrently. Therefore, one of the main policy challenges that will need to be faced in the medium and long run is the problem of how to manage congestion in unlicensed spectrum. We examine three approaches for alleviating the congestion problem: licensing new bands of spectrum, allocating additional unlicensed spectrum bands, or developing new technical solutions which can more efficiently utilize available spectrum. (This is part of the motivation for the research projects we have engaged in exploring technical solutions to coordinating competing access to the wireless spectrum by different access providers.)

To learn more read: Residential Resale of Wireline Broadband via Wireless

Characterizing application and network services requirements and capabilities

Existing wireless networks provide a wide variety of service capabilities. Due to the inherent nature of wireless transmissions, these services are often characterized by high error rates, variable bandwidths and delays, and unpredictable interruptions. Users and applications are somewhat adaptive in their ability to handle these variable service conditions. However they are not completely flexible nor does the user perceived quality vary in a uniform fashion with the changes in network service.

Traditional QoS parameters often poorly match user preceived notions of quality in wireless networks. This research therefore examines two qos parameters that characterize application and network service requirements and capabilities. These parameters enables the translation of user perceived notions of application utility into appropriate network service selections and controls. In particular the application utility specifications capture important time constraints and adaptation paths that have not been adequately expressed previously. We demonstrate the benefits of our model by showing how applications and wireless networks can improve the quality of service delivered.