CS 294-7, Special Topics: Wireless Communications and Mobile Computing
Draft Project Ideas, Spring 1996

Evaluation Tools and Methodologies for Wireless Networks

Mobility Traces

There are few published studies of user mobility. This is due, in part, because there are few operational mobile systems from which to collect data and in part because those systems that do exist keep the data secret for fear of revealing user identity (or competitive weaknesses).

It would be a major contribution to collect and analyze detailed traces of user mobility. Since Soda Hall is not fully covered by wireless access, this presents some real challenges! Here are some strategies that could be pursued. First, you could ask a group of test subjects to log their locations in a notebook for a few days or a week. Second, you could examine the security tapes for Soda Hall to measure the distribution of people entering and leaving the building (you should also examine the class schedules and the reservation lists of rooms for informal meetings, qualifying exams, etc.). Third, you could develop a laptop application with detailed "clickable" maps of the floors of Soda Hall (these are already available on the Web), and ask some subjects to carry it around for a day, using the application to capture his/her location throughout the day.

Here are just some of the questions that could be asked about such traces. How often do users reposition themselves? Are there a small number of standard routes that a given user commonly follows through the building? What is the density of users in the building, and how does it change as a function of time or day of the week?

Wireless Network Benchmarking

We have wireless networking software that has the ability to collect packet traces and measure physical network performance parameters like received signal strength and bit error rates. We need to design benchmarks (for example, a rate controlled UDP traffic generator) and to collect and analyze traces from several different networks (in-building wireless LAN, Metricom packet radio network, CDPD cellular data network, Hughes DBS satellite network, etc.). It would be great to collect as large a library of (reproducable) benchmark traces as possible.

Network Simulator

Rajiv Bragrodia's research group at UCLA has developed an event-based simulator called Maisie, and they have developed channel models for investigating a variety of packet radio routing algorithms. We would like to extend this simulator for one or more networks of interest, such as WaveLAN, Metricom, CDPD, or Hughes DBS, validate it against the real network to the extent possible, and use it to investigate new transport and routing algorithms.

Mobile Routing

The Mobile IP specification can be extended in a variety of directions, any of which would make an interesting project:

Overlay IP

Mobile IP was never designed to support routing in overlay networks. The needed capability is for the HA to choose among alternative wireless networks for providing connectivity to the Mobile Host. How should the Home Agent be modified to allow it to select a given network interface (and IP address) for a particular kind of network traffic? What kind of session or stream abstraction is required? How is information exchanged between the Home Agent and the Mobile Host to establish the policies by which a route is chosen?

Hierarchical Routing

A potential limitation in the current Mobile IP specification is that the Home Agent needs to be informed about every change in location of the Mobile Host. These could be high latency operations if the mobile is far from home. One possible improvement is to introduce hierarchical routing authorities (Foreign Agents) to which the HA forwards packets, which are then reforwarded locally to FAs associated with individual microcells. Most location updates are thus handled locally. The details of this scheme need to be worked out and evaluated in depth (note that mobility traces would be of great value in this evaluation).

Mobile Network Management

Cell Admission Control and Airlink Load Balancing

Our current implementation of Mobile IP does no admission control and makes no attempt to do load balancing between cells or alternative networks. How should the registration process be generalized to provide these capabilities? How does it interface with the HA to assist in making routing decisions?

Mobile Security and Privacy

Our current implementation of Mobile IP does nothing to protect user identity or provide end-to-end encryption. It would be worthwhile to design and implement either a public key or a shared secret/challenge-response prototype system. Note that we might be able to obtain some Tessera cards for use in this project.

Internet Firewalls and Mobility

Many organizations deploy Internet firewalls to restrict access to external users of resources inside their corporate networks. Develop and evaluate some schemes that could allow local users who are roaming in foreign networks to access resources inside the firewall. In principle, the problem is the same as that of users who need to dial-in to a corporate network from outside. Such systems typically make use of secure terminal servers inside the firewall. How might a local authentication agent be constructed to provide adequate security for external access?

IP Spoofing and Mobility

Routers with IP Spoofing detection enabled cause many problems for the existing Mobile IP specification. The source address of the packet is the MH's address in its home network, but the router interpretes this as an attempt to spoof the Internet by providing an incorrect IP address on purpose (the router will refuse to pass the packet!). Propose and evaluate methods for circumventing this problem.

Wide Area Roaming Architecture

Mobility software has not as yet been widely available. Most research groups working on Mobile IP have implementations that are only operational within their own local administrative domains. What will it take to implement a true wide area roaming architecture around the Mobile IP specification? Most of the issues are related to resource allocation (local users should be given priority access to network bandwidth, compute and printer resources, etc.) and accounting (e.g., keeping track of resource usage in the visited environment; this can form the basis of a credit system when mobiles from the visited environment become visitors in turn).

Reliable Transport in Wireless Networks

One of the topics we will be covering this semester is how to improve reliable transport performance over wireless links. This is crucial for supporting the "killer application" of the 1990s, namely Web Browsing. A number of schemes have been proposed, such as local retransmission of packets ("snooping"), split TCP transactions, fast retransmission timers on the wireless link, etc. All of these strategies have been evaluated in the context of wireless local area networks, which typically do not implement reliable links through error corrections or link layer retransmissions. How well will these mechanisms work with wide area networks, such as multi-hop packet radio, satellite networks, or other wide-area wireless networks that implement more reliable link technology? What other mechanisms, such as adaptive retransmission timers or explicit identification of lost packets, might also help to improve TCP performance over these other kinds of networks?

Direct Broadcast Satellite Technology

Design and Evaluation of Applications for DBS

Directed Broadcast Satellite technology is inherently asymmetric, i.e., the downlink bandwidth is much greater than the uplink bandwidth. There is much interest in investigating applications that are a good match for the asymmetric nature of the satellite link. For example, conventional database systems pull data from the database system in response to queries. But the satellite channel is better suited for pushing data over the satellite, where end users tune and filter the data they are interested in. There are aspects of the DBS system that would be interesting to investigate in more detail, such as the way transmissions are encrypted for multicast groups and the ways that users can join and depart from such groups (thus requiring a change to the group's encryption key).

Some possible applications include subscription-based services (e.g., periodic transmission of stock market ticker, weather and traffic reports, etc.) or satellite distribution of the most popular 1000 (5000, 10000) World Wide Web pages (it would be interesting to couple this with the Inktomi search engine, which can keep statistics of which pages receive the most frequent hits). A project like this would probably involve a simulation and an analysis. The DBS system itself is probably too closed for actually implementing a service like this.

Network Design

There is little in the way of well known literature on the end-to-end performance of TCP/IP networks over satellite constellations. It would be interesting to investigate routing and reliable transport issues for low earth orbiting satellites. This study would involve developing a simulation model of the satellite channel and some other aspects of satellite constellations, and then perform some analysis of network performance.
Randy H. Katz, randy@cs.Berkeley.edu, Last Updated: 29 December 95