As mentioned earlier, several new challenges are posed by the urban wireless grid network. To address these challenges, five key tasks are defined below. It will be noted that some of these tasks are open ended and in fact are more suitable for a multiyear effort than a single year project like the one proposed here. We will attack and define solution approaches for all the tasks. However, we will defer completion of the more complex tasks to future extensions of this project.

MAC layer enhancements; QoS Support

The traffic scenarios depicted in the previous section exhibit an interesting mix of connection oriented (voice, video) and connectionless (best effort) traffic, fixed and mobile terminals, and of "centric" vs. multihop network configurations. In the context of the above defined scenarios we will compare various MAC layers. These will include IEEE 802.11, and proposed evolutions for QoS [802.11], [Har00], [Hart98], [TaG97], [HaD99]. Moreover, the IEEE 802 standard committee is expected to create new working groups for Mesh Networks and for Vehicle Networks. Since our sponsor, ST Microelectronics, is an active participant in the IEEE 802 Working groups, we will leverage the ST connection to keep abreast of any draft recommendations issued by IEEE regarding Mesh and Vehicle networks and will include them in our evaluations. More generally, we will propose and evaluate new extensions to current MAC protocols capable to effectively support the needs of advanced applications in a wireless mesh network. To start, we will engage in the following MAC activities:

We proposed in a previous project a novel MAC layer for Wireless LANs, named Interference Aware MAC (IA-MAC) [CeM03], [MaC03] which extends the capabilities of the basic IEEE802.11a/b/g in environments with high interference, both in wireless LAN and ad hoc configurations.

This requires minimal modifications to the IEEE802.11b MAC DCF in order to make multiple parallel communications feasible in a scenario where interference is not negligible.
We propose to modify the physical Carrier Sensing by setting a high value of capture power. This way, each node recognizes the channel as busy only if highly overloaded. Moreover, we propose to slightly modify the CTS packet to include information on the Signal-to-Interference Ratio (SIR) measured by the node returning the CTS sender, say node B. Namely, node B, receiving a RTS includes in the CTS the ratio of the power received on the RTS and the perceived interference level. All the mobile nodes within node B range with own packets to send, upon overhearing the CTS decide whether to start transmitting their packets or defer on the basis of the SIR information coded in the CTS. More precisely, they refrain from transmitting if they estimate that their transmission would cause node B SIR to fall below a specified threshold.
IA-MAC offers a partial solution to the exposed terminal problem, granting a good gain with respect to the basic MAC mechanism. We evaluated IA-MAC in ad hoc and wireless LAN configurations via simulation using QualNet simulator [Qualnet].

We will study the protocol in scenarios where the transmitted power is not uniform over all nodes. Furthermore, we will work on two future improvements to IA-MAC; the former, by choosing whether to set NAV also after the reception of a RTS packet, taking into account the transmission durations; the latter just avoiding some collisions on the ACK packets (also in this case, exploiting the transmission durations).

We worked on novel system architecture for QoS support in WLAN. We focused our work on IEEE 802.11 family but the concepts introduced are fairly general and applicable with minor adaptations to several different technologies, such as Bluetooth, UP-5, Hyperlan, etc. There are several concerns about the QoS support in the currently leading WLAN standard. The IEEE802.11a, b, and g, version do not have any QoS support unless extended with the IEEE802.11e. The QoS support in the IEEE802.11e still presents several open issues to be addressed in this study:

  • Traffic Categories assignment: The standard does not define how the Traffic Categories (TC) are assigned. They can be assigned directly by the application or by another entity in the network although an application update, in this case the introduction of a middleware component may be required. If the network is already QoS enabled (i.e. DiffServ) the proper TC can be assigned mapping the upper layer QoS parameter. In general edge-oriented technique is more appropriate since it cannot take into account specific media related constraints.
  • Content Based Policies: Within the same flow each packet may have a different level of importance with respect to the perceived quality; in the case of MPEG2 video, for example, I-frames are much more important than B and P-frames in terms of perceived quality. Content based differentiation may be the key in a networking scenario with high multimedia content such as home networking.
  • Hybrid Networks: The deployment of IEEE802.11e will probably require a hardware substitution or a firmware update depending on the manufacturer. During initial phases of deployment, the nodes that use different technologies (802.11b and e) may well be in the same network (being served by an 802.11e enabled Access Point) and such situations may exacerbate unfairness between 802.11b and 802.11e clients.

Regarding the above efforts, we worked in the past on a cross-layer QoS support for IEEE802.11x, (a,b,g,e,n); mainly we introduce a classification policy point that uses layers 4-7 information present in the packets to perform a flow classification and further packet type classification within the same flow. This allows us to apply different layer 2 strategies based on the content of the packet. Similar approaches have been partially described for the Bluetooth environment in [KaC03]. In [Her02] a content weighted retransmission policy for MPEG4 over IEEE802.11b is also introduced. The proposed architecture uses a Wireless Quality Enhancer located at the wired backend to classify the different traffic and define an appropriate MAC layer policy set at the base station. The basic idea behind is to use the layer 2 customizable parameters to improve the QoS perceived by the user. In this project, we will apply these techniques to the emerging MAC protocols and applications in the urban grid environment. While the architectural design is fairly general, the simulation and implementation will be oriented to the IEEE802.11 family using the OpenAP [OpenAP] technology and an US Robotics 2450 AP [UsRob].

A hybrid simulation approach will be also used to verify the architecture scalability. The proposed scheme rests on the concept of a MAC layer that adapts its behavior to the transport and application layer requirements. This requires a per flow MAC policy that should accommodate network, transport, application and media requirements for an optimal user experience.

Horizontal and Vertical Handoff

In a global infrastructure we can identify two different classes of network interconnections, namely the vertical interconnection from one type of network to another (i.e. from a MANET to the wired Internet or from a PAN to sensors and actuators) or the horizontal interconnection between MANETs at the same level. The vertical interconnection from sensors to PANs, MANETs and the wired Internet, represents probably the most novel aspect of the proposed research. In a car-to-car system we have wireless communications among all the components of the car, and communications to the Internet and to other cars that are the most innovative applications. So the mobile user can switch his network access from one to the other. This problem has been extensively studied and deployed in the cellular systems, and it becomes increasingly important in the computer networks nowadays.

Figure 2: Soft and Hard Handoff

Differing in the number of connectivity to the base stations, the handoff can be characterized into hard handoff and soft handoff [Pol96], as shown in Figure 2. Hard handoff means that only single connectivity to one base station is allowed at all time for the mobile device, and it is widely used in the current GSM and GPRS systems. A soft handoff, on the other hand, allows two or more connections to different base stations to be kept, i.e. the network is overlapped, and it is used in the newly CDMA based cellular systems, such as WCDMA and 1xRTT (cdma2000).

Figure 3: Horizontal and Vertical Handoff

Additionally, based on the number of network interfaces involved in the handoff, the handoff can be characterized into vertical handoff and horizontal handoff [StR98], as shown in Figure 3. A vertical handoff is a handoff between two network access points, which are usually using different network connection technologies. For example, when a mobile device moves out an 802.11b network into a GPRS network, the handoff would be considered a vertical handoff. A horizontal handoff is a handoff between two network access points that use the same network technology and interface. For example, when a mobile device moves in and out of various 802.11b network domains, the handoff activities would be considered as a horizontal handoff, since the connection is disrupted solely by device mobility.

For the urban grid application, a universal seamless handoff solution is desired in order to provide continuous connectivity even when the mobile switches from one access point/medium within seconds. The handoff may happen either vertically or horizontally; however, the proposed handoff solution should guarantee the uninterrupted services for the mobile users. Additionally, the deployment of the handoff solution should minimize the changes of the current Internet infrastructure. To achieve these criteria, we intend to take advantages of the existing NAT and IP Tunneling techniques to create a seamless handoff testbed. Differing from other well-known handoff solutions, our proposed solution will follow the middleware design philosophy [FoG97] so that it is scalable and also easy to design, deploy, and extend.

Moreover, in our proposed handoff architecture, three levels of adaptation will be added to enhance the service performance in the mobile scenarios. The first level of adaptation is on the mobile hosts so that mobile hosts are granted the ability to decide the most appropriate time and target network interface to handoff. The handoff decision can be based on various concerns, such as link capacity, energy consumption, current battery status, and connection cost. The second level of adaptation is on the middleware so that it can trans-code "on-the-fly" the data to fit the new network conditions after handoff. The third level of adaptation is on the Internet servers so that they can become aware of the handoff events on the mobile hosts and adapt their own service rate (TCP and UDP sending rates) based on the new network conditions. These three adaptations will enable better services, and will lead to more efficient network resource utilization.

TCP in wireless/wired environments

Shifting now to the transport layer, a critical issue is TCP performance when the mobile user is on an IEEE 802.11 access LAN and receives a large file from a remote Internet server. Current TCP implementations suffer of severe degradation if the IEEE 802.11 link is noisy due to the difficulty in distinguishing between packet loss caused by congestion or wireless interference. We have proposed MAC layer remedies, which have shown promise but require MAC firmware changes. We will study the impact of those changes on the vehicle urban grid scenario via simulation. Any promising results will be promptly communicated to IEEE 802.11 standard committees in the hope of driving the MAC standards to TCP friendly solutions. We also plan to continue explore with the implementation of "TCP friendly" network layer protocols. The latter are independent of the MAC and much easier to implement in a LINUX kernel. Given the importance of TCP based applications in the urban grid environment, we expect these results to be an important contribution of this study.

Sensor platforms

In the urban grid of the future we envision cars to be equipped with various sensors (video, chemical, vibration, radiation, shock, etc) and thus become true sensor platforms. Naturally, the type and degree of sensorial support will vary from private cars to public and emergency response vehicles. In this environment "virtual organizations" will be created (e.g. police cars, taxi cabs, ambulances, etc) at different times to attend to different missions. These organizations which extract fuse and maintain different subsets of sensor information collected by the car platforms. Distributed directories will be maintained in part in the wired internet (for example, using existing hashed based directories, eg Pastry, Chord, CAN, etc) and in part in the ad hoc net (via epidemic dissemination, say). These needs can be met in a number of different ways. In our study, we plan to attack this problem using a mobile Grid computing model.
The Grid computing model recently introduced for fixed resources [FoK01][Fok99][FoK02] (e.g. supercomputers and physics applications) lends itself well to large-scale dynamic resource sharing and transient associations of devices forming virtual organizations. The OGSA (Open Grid Services Architecture) has standardized the concept of grid-services. Every resource on the grid is viewed as a service and devices in the same virtual organization can access a shared resource securely. This architecture fits very nicely with the application scenario of dynamic associations of police cars, ambulances and even civilian cars. One visionary application could be a high-speed chase on the freeway. Video-cameras mounted on police cars, buses or even futuristic civilian cars would be hosted as grid-services. Police cars authenticated properly would be able to form virtual organizations accessing these video feeds to get precise knowledge about the movements of the target vehicle. Another exciting way for the Grid model to enable such communications is to use the grid-services architecture to form virtual organizations at the grid-layer and then set-up the routing at the network layer through this "application layer awareness" to optimize the communication paths. We propose to investigate this idea by building grid-services which will interact with the network layer to setup the routing parameters depending on the MAC layer technology being used. In this way, the network will be fully-reconfigurable via the Grid, not only in terms of which nodes (cars) to include in a virtual cluster, but also which routing protocol and which physical technology (e.g., 802.11 or GPRS) to use for communication itself.
Coordinating a dynamic set of fast-moving resources requires an accurate monitoring and resource lookup service. The current Grid based MDS (Monitoring and Discovery Service) [CzF01] consists of a hierarchical organization of indexing servers which does not lend itself well to a highly mobile service scenario. An investigation into alternative ways of storage and lookup of content and services had led to very promising results. A key result was that organizing data in a distributed manner much like a Pastry network, and using multiple paths to increase reliability can lead to much better performance in highly mobile networks (like dynamic associations of cars). We shall also investigate the use of ad-hoc network mechanisms like epidemic dissemination in conjunction with PASTRY Pastry like approaches to get a hybrid solution to the problem. For instance, Pastry could be used on the wired Internet part of the storage and lookup network, while ad-hoc epidemic dissemination or gossip kind of approaches could be used between the cars when infrastructural support is lacking.
One of the most interesting features of vehicular networks is the sporadic presence of the infrastructure network. There could be stretches of urban areas with IEEE 802.11 hotspots with very good connectivity followed by rural areas with hardly any infrastructure support. Following a grid-model could facilitate transitions from a peer-to-peer communication model to an infrastructure-enabled communication model and then back to a p2p model seamlessly. Thus, the car network could use the wired Internet as well as the shared car system for storage and lookup depending on availability and capacity.

Mobility assisted routing

In an ad hoc network node mobility is generally the "adversary": network protocols are designed exactly to protect user performance from mobility. There are, however, ways to exploit mobility and dramatically increase capacity in large ad hoc networks. We mentioned information kiosks before as one such example. Another way is to allow servers to opportunistically piggyback information about their service and location on passing users. At the other end of the urban grid, a client, by probing neighbors, can efficiently learn the position of the nearest server without engaging in a costly broadcast search [Sul00][BlL02]. This feature is particularly important for delay sensitive applications.
We plan to examine and evaluate this motion assisted advertising and routing strategy. In fact, we expect that in the future, various franchises/establishments will advertise their presence with electronic beacons (instead of unaesthetic neon lights!) in order to improve the urban landscape. Resource discovery (e.g., find the nearest movie theater showing a particular movie) can be carried out using a search on the urban grid to locate the nearest beacon. A more efficient approach is to have cars promiscuously listen to all sorts of advertisements emitted by the environment and "store" them in memory. Some cars will thus capture the beaconing object ID (along with timestamp) and carry it along for a ride (until timeout). A car interested in a particular movie, say, will then query passing cars to see if they have heard the movie beacon. Upon getting multiple positive responses, the car will choose the response with the most recent timestamp and proceed to drive in the desired direction (which will lead to the nearest theater). Alternatively, the car can route a packet to the beacon source by using a focused search that threads backward hopping from car to car to the beacon. This "mobility assisted discovery" scheme is an extension of the FRESH routing scheme recently reported in [FeG03]. We intend to assess via simulation the impact of this type of "mobility assisted" resource discovery and routing protocols in our urban grid environment.