Mobility in Wireless Networks NSF Workshop Rutgers, July 31-Aug 1, 2007 Mario Gerla (UCLA) Dipankar Raychaudhuri (Winlab)

2 The Mobile Internet New generation of powerful portable devices: –Can support most Internet needs Wireless speeds growing constantly: –4G expected to achieve 40Mbps –WiFi up to 100Mbps Opportunistic ad hoc networking facilitates P2P applications

3 The paradigm shift Traditional wireless mobility: –Last hop connectivity –Soft handoff (horizontal, vertical) –Most data and services still in the wired Internet –Advanced ad hoc networking only in tactical and emergency scenarios

4 The paradigm shift (cont) Emerging Wireless, Mobile Internet –The data is collected by portable devices, and may stay (and be searched) on the devices for a long time: Urban sensing (vehicle, people) Medical monitoring, etc –This creates new challenges Distributed index (ie, publish/subscribe) to find the data Data sharing among mobilers via opportunistic P2P networking –Infrastructure used if more efficient than pure wireless: “Reconfiguring the Infrastructure” Privacy, security, protection from attacks Intermittent operations (mobile nodes can become disconnected); delay tolerant applications; disruption tolerant networks

5 Examples Car Torrent Blue Torrent Medical monitoring CarTel EU Haggle project Diesel Net Etc…

6 Bluetooth: Content Sharing A A B C B C DD A D B C

7 Collection Center ZigBee Intra-PAN link ZigBee Enabled PDA ZigBee Medical Sensors ZigBee Inter-PAN link Patient Collaborative Health Monitoring: ZigBee as Health-Net

8 Nurses Cache-and-Forward Patient Data to Doctors

9 Objectives of the NSF Workshop How does “mobility” change traditional network architecture and design? –Applications –Protocols –Mobility models What new research is needed to make progress?

Emerging Mobile Applications

11 A New Generation of Mobile Applications Distributed Integrating heterogeneous infrastructure (e.g., WiFi, cellular, satellite) and ad-hoc networking Location-aware –Opportunistic, predict, control Exploit mobility –Homogenous or heterogeneous mobility –Individual or swarm mobility User behavior-aware Location privacy sensitive Self-configurable, self-tunable, remotely manageable Energy-aware

12 Mobile Application Examples Vehicular applications –Safety, traffic information, route planning Content-sharing applications –Entertainment (video, audio), games Mobile external sensing –Urban pollution sensing, accident reporting Mobile ad-hoc services –Relaying to near-field users Emergency applications –Disaster recovery Mobile network management Mobile social networking –Mobile Facebook

13 Research Challenges Mobile application design –Location-aware, exploit mobility, gathering feedback and traces Performance and QoS –Delay tolerance, channel variations Cross-layer communication design –Exploit mobile application context information Security issues –Location validation, Location privacy, Trust management Social aspects –Event-driven and event-generated mobility

14 Research Challenges (cont’d) Mobile data management Fault tolerance –In the presence of mobility Remote maintainability –Deployment, configuration, upgrade, debugging Application and service-oriented protocols –Over mobile networks

Robust, Motion Resistant Protocols

16 Main Message Changing the view on mobility: –Mobility has become an integral attribute of the Internet and we need to design for it. –Without mobility support, the Internet cannot be invisible. There is a big gap between the opportunities that mobility enables and the practical protocols that can take advantage of it. Design for mobility requires a clean-slate approach to communication protocols in wireless networks and the Internet Design for mobility has direct implications on the Internet design, in-network storage and localization information being key factors Standards are needed for benchmarks.

17 Why Is Mobility Important? (1) Mobility impacts: –the conditions in which protocols must operate, –the state and context that nodes can use to communicate, and –the problems that protocols must solve. Examples: –The state of links is a function of mobility (e.g., link lifetime, fading, multipath effects, direction of a link, etc.) –The neighborhood of a node changes with mobility, which impacts reliable exchanges, channel division (space, time, code, frequency) among neighbors, and forms of cooperation between senders and receivers (e.g., virtual MIMO, network coding) –End-to-end paths change with mobility, which impacts path characteristics (in-order delivery, delay, throughput, lifetime of paths, etc.) and the allocation of resources over paths to satisfy application requirements.

18 Why Is Mobility Important? (2) Examples: –Supporting security is more difficult with mobility: Identities of trusted nodes must change –Privacy can be compromised with mobility: Example: Node can be tracked by the perceived location of its transmissions –Layering of protocols is hard(er) with mobility –Policy-based dissemination of information is more difficult with mobility (dynamic routing with multiple constraints) –Feedback to control data rate is not as useful if paths change (bottleneck state is stale) –Some form of dissemination is enabled or simplified with mobility.

19 Designing for Mobility Some protocols benefit from mobility: group mobility, etc. –Use mobility as a mechanism for information dissemination Controlled mobility: –nodes move around to improve topology, deliver data, store-carry-forward, –trajectory planning and changing what routes Interest-driven “physical” dissemination: –How should opportunistic data mules handle data? Content-driven routing

20 Designing for Mobility: A Clean Slate Exploit broadcast nature of links & in-network storage OSI/TCP architecture is no longer “the best”…What is the new slate? –MAC issues: MAC should work on broadcast and directional transmissions; support many-to-many rather than one-to-one communication –Network issues: Naming (no need for addresses?), attribute- based queries, geo-location is important, resource discovery (no DNS) –“Beyond routing”: resource discovery replaces route discovery; need for binding of resources/services on the basis of names;

21 Designing for Mobility: A Clean Slate (cont) –Route binding: “Opportunistic use of resources” cooperative x-mit schemes (take advantage of gains at PHY) incentive mechanisms (battery life, use of spectrum), cooperation using memory, virtual MIMO Use mobility of nodes to cooperate as data mules; need for coordination to decide which nodes move where Peer-to-peer opportunistic transmissions: how to cache, how to route, post box (beyond foreign agent). Trade storage for retransmissions –State information vs opportunism: Use of CSI induced by mobility to help cooperation vs opportunistically forwarding w/o CSI. –Tolerance to various forms of disruption (e.g., no connections)

22 Changing The Internet Design Use of storage and location information must be considered in the global routing design. Use of location information: IPv6 can be used but we must find anonymous location information in addressing…use of proxies and in-net storage. Easier approaches are above IP. Privacy and security implications There are localized protocols that can (should be) used (e.g., a DHCP-like approach) Mobility creates a stronger focus on security: –We do not know the local neighborhood. Opportunistic mobile routing infrastructures will become important Mobility changes the expectations for services (anywhere, anytime), but maintaining performance with seamless mobility is difficult.

23 Role of Standards What should be standardized? –Benchmarking different protocols (modeling and otherwise) –Understand dynamics of system to understand what to standardize –Need for a reference model capturing connectivity structures/motion patterns and spanning different scenarios and protocols. –Parameters that should be of interest are open –Look at the “connectivity structure” of a network (dense or sparse, guidelines)

24 Multicast Coupled with mobility, maintaining routing structure is hard Using m-cast to satisfy localized services Potentially more overhead that unicast In-network storage can help, but the consistency of data is an issue Need to study one-to-many-to-one (convergecast + multicast + anycast) Pull vs push with in-network storage

25 “Things to Things” Communication Location is key aspect of interaction Mobile sensors (e.g., health monitoring) with limited power need protocols Internet at the service of users and things that move.

Mobility Models and Mobile Testbeds

The Models Motion, Traffic, Network

28 Model Flexibility Multiple scale models –Micro and Macro levels, (e.g., from stop signs to cross town patterns Multi-faceted scenarios –Combines motion, data traffic, map, infrastructure –Interrelation between data/motion; data caching; aggregation, etc Trade off between accuracy and usability –Different applications may focus on different parameters

29 Traces to Models Traces: –Lack of cellular traces (owned by providers) –Lack of vehicular traces (not enough testbeds) –Lack of social network tracing experiments –Scarcity of urban traces Interplay/synergy of: –Measured traces –Synthetic models/traces –Theoretical motion/traffic models

30 Metrics and Parameters Motion Impact on Data Performance: –How are the metrics impacted by the particular motion patterns, –How do the motion patterns impact the traffic, Consider new “mobility” measures: –Inter-contact time, neighborhood change rate, partitioning, clustering, etc –Ideally, a few motion “primitives” that can cover most scenarios and allow cross comparison of test experiments

31 Performance Benchmarks Well defined benchmarks Knobs, ie, flexibility –Need to understand effect of knobs –Need knobs tunable to applications Sound design methodology –Ability to tradeoff accuracy, complexity etc –Verifiability

32 Evaluation Methodology Guidelines for community Model validation Model implementation verification Sound statistical analysis of results

Testbeds

34 Testbed Flexibility Multi-layer multi-user vs. single layer/user testbeds Heterogeneous (hardware, protocols, applications) Broad range of motion patterns: –From pre-scheduled to controlled and spontaneous Broad Range of devices: –From small scale testbeds (motes) to large scale testbeds (vehicles)

35 Testbed Scalability Testbed expansion with simulation and emulation Integration with real world networks and applications

36 Testbed Realism Need more than what simulation already gives us!! Need to understand: –Realistic user behavior in reaction to motion, data etc –Realistic channel behavior –Real implementation/HW constraints Uncover: –interactions between layers and inefficiencies –Incorrect common beliefs Appreciate: –HW, SW, Mgmt costs

37 Measurement Methodology Guidelines for the community Validation of the model (motion, traffic, etc) Verification of the implementation Repeatability Sensitivity analysis Sound statistical analysis

38 How can the Internet (and GENI) support mobile applications? Addressing and routing –Geo-routing –More generally, attribute based routing –Mobility support Interaction with the infrastructure –Off loading the wireless internet Wireless as emergency network –When the infrastructure is brought down Congestion control assistance Security, protection against attacks

39 Role of GENI Wireless Mobility testbeds Vehicular networks People to people networks Small scale - augmented by simulation (hybrid) Interoperation etc