Vehicle Networks Networking Layer: Routing Univ.-Prof. r. Thomas Strang, ipl.-inform. Matthias Röckl
Outline Introduction Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Classical networking Ad-hoc networking Proactive routing Reactive routing Vehicular Ad-hoc Networking Requirements Geographic routing
Introduction efinition Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 With the wireless communication specified in the WAVE (Wireless Access in Vehicular Environments) standard, we now are able to communicate between a source node and one or more destinations within the direct communication range (~00-000m) Some applications require a larger distance (e.g. traffic jam warning) to be effective A solution is multi-hop communication, i.e. the transportation of a packet via several hops, for the delivery of packets to one or more nodes located beyond the single-hop communication range Multi-hop communication is one of the main task of the network layer according to the OSI reference model Terminology: Receiver Sender Source S estination Transit node
Introduction Flooding Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Simplest solution is Flooding: source broadcasts packet, every recipient re-broadcasts the packet, etc. Problem: High bandwidth consumption High redundancy Broadcast-Storm problem: All receivers S of a message try to re-broadcast the message at nearly the same time High bandwidth consumption arises for every single packet to deliver Idea: If the same transmission path can be used several times, a reusable path can be established
Classical networking Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009
Classical Networking Routing Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Routing = classical networking in static wired/wireless networks Establishment of a suitable, reusable route between source and destination node prior to packet forwarding Tasks: Route iscovery: Find a route from a source to one ore more destinations based on a search algorithm Route Maintenance: Maintain routes during runtime (re-setup routes in case of topology change, setup alternative routes in case of broken links, etc.) Route Usage: Forward the data packet along route Optimization: Maximize or minimize some metric for the path (e.g. minimum number of hops)
Classical Networking Characteristics Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Stable links iscovery and maintenance of static routes Route adaptation mainly in access nodes of the backbone Examples: Fixed-line internet, mobile internet, home network S End system Wired/Wireless Connection Backbone Wired/Wireless Connection End system
Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Classical Networking istance-vector Routing Each node maintains a forwarding table (= distance vector) with the next hop on the shortest path for all destinations in the network, i.e. a node does not maintain a complete topology graph Each node periodically advertises its distance vector to its direct neighbors Example: Routing Information Protocol (RIP) used for instance in ARPANET and IPv4 (RFC 08) Node is merely aware of the distance (hops, delay, etc.) to every destination Algorithm based on istributed Bellman-Ford s search algorithm for single-source shortest paths in a weighted graph (breadth-first search) isadvantages: Long convergence times (i.e. time until a consistent view on the network topology for all nodes in the network is (re-)established) Count-to-Infinity problem: recursive false updates inducing routing loops due to limited topology knowledge A C B E F istance Vector at A estination Hop Next istance B B C C C 4 E C F C 7
Classical Networking Link state routing Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Every node maintains a complete network A B F A B topology graph C E C E Every node periodically sends the state of the links to all its neighborhood to its neighbors Algorithm based on ijkstra s algorithm for the single-source shortest paths in a weighted graph (uniform-cost search) A B F Example: Internet routing protocol Open Shortest Path First (OSPF) (RFC 740), E often used instead of RIP Advantages: C Better convergence time Better loop prevention Possibility of directly selecting an alternative route in case of a broken link isadvantage: Excessive control overhead in case of unstable links, e.g. due to mobility Requires more storage space to maintain the entire topology information F
Classical Networking Theoretical view Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Topology-based: graph G(V,E) with V = set of nodes {v, v n } and E = set of (wireless) links, (v,v ) E V x V w(e) = weight function, w: E R Weight of path p = v v v k is Routing: Finding an optimal path (according to a costs metric) from source S V to destination V within the graph G by traversing a number of edges e E. This can be achieved by graph theoretical search algorithms: Breadth-first search (Bellman-Ford): Expand source node first (= inspect all outgoing links), then expand every successor of the source node, then their successors, etc. (see lecture Algorithms and atastructures, e.g. in SS 006) Uniform cost search (ijkstra): Instead of expanding the shallowest node, expand the node with the lowest path costs (see lecture Advanced Topics in Intelligent Systems, e.g. in SS 008) k wp ( ) = wv ( i, vi+ ) i=
Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Classical Networking Breadth-first search A C B E F A C B E F A C B E F A C B E F A C B E F A C B E F A C B E F A C B E F 4 6 6 7
Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Classical Networking Uniform cost search A C B E F A C B E F A C B E F 4 A C B E F A C B E F A C B E F 7
Ad-hoc networking Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009
Ad-hoc networking Overview Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Wireless network without necessity of fixed infrastructure Nodes are randomly distributed and may change their location Examples: Wireless mesh networks Wireless sensor networks (WSN) Mobile Ad-hoc networks (MANET) Application fields: isaster detection (fire, landslip, flood, etc.) Agriculture & wildlife observation Smart environment control (home, office, etc.) Military tasks (battlefield surveillance) Multi-hop communication mainly based on refinement of classical routing algorithms: reactive route establishment in contrast to proactivity in classical networking Telos mote Wireless sensor node ZebraNET
Ad-hoc networking Characteristics No pre-defined links ynamic discovery of routes S Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009
Ad-hoc networking Proactive Routing Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Each node continuously maintains up-to-date topology information independent of communication activities Each node stores topology information in a local table (table-driven) Periodic updates of neighborhood table even without data communication (control traffic only) Every time the network topology changes, control packets including route information have to be exchanged Superior in case of only small and seldom changes in the network topology (e.g. randomly distributed static sensor nodes for agricultural observation) High control overhead in case of frequent changes in the network topology not scalable in highly mobile ad-hoc networks Example: estination-sequenced istance Vector Routing (SV)
Ad-hoc networking estination-sequenced istance Vector Routing (SV) Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 istance Vector algorithm for ad-hoc networks Route calculation based on istributed Bellman-Ford Every node stores the distance (i.e. costs) to every destination together with the next hop in its distance vector Every node proactively broadcasts its distance vector to its neighbors Full dump: sent periodically, includes whole distance vector Incremental update: sent on minor topology changes, includes only topology changes since the last update Upon reception of a distance vector the receiver compares the distance to the destination with the value in its own distance vector. If the distance is smaller than the own distance plus the costs between sender and receiver, it is substituted for the newly received distance
Ad-hoc networking Reactive Routing Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Nodes only initiate route discovery when required Route discovery just before data transmission (no unnecessary periodic table updates, but may cause extra delay) Neighborhood information is only sent if required (on-demand), i.e. a packet has to be routed and no route has already been established Caching of routes possible decreases control overhead and route setup latency Sequence numbers in packets avoid routing loops and allows transit nodes to detect retransmissions Superior to proactive routing if links are unstable (e.g. networks with high mobility) better scalability in highly mobile ad-hoc networks Examples: Ad-hoc On-demand istance Vector Routing (AOV) ynamic Source Routing (SR)
Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Ad-hoc networking Ad-hoc On-demand istance Vector Routing (AOV) SV derivative with reactive route setup Algorithm:. Source floods the network with Route Request (RREQ) packet including source and destination I. Every transit node stores the source I and the sender of the RREQ in its routing table (backward learning). If the transit node knows the route to the destination, it sends a unicast Route Reply (RREP) back to the destination via the optimal reverse path (-e-c-b-s) 4. Every intermediate node in the reverse path stores the destination I and the sender of the RREP in its routing table (forward learning). When the source eventually receives the RREP, it transmits the data to the sender of the RREP, Routing table at Node c: After RREQ: (backward learning) est Next ist S b After RREP: (forward learning) est Next ist S b e Optionally AOV allows to periodically exchange hello messages in order to detect broken routes if neighbors are out of range d a t t c S 7 6 8 b e 4 RREQ broadcast at time t RREP via reverse path at time t
Ad-hoc networking ynamic Source Routing (SR) S Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Source Routing: Source determines complete path to the destination when forwarding a packet (path information included in the packet) ynamic Source Routing: Similar to SV but RREQ and RREP contain path information including every transit node (route record) Route record is reversed by the destination and sent back in reversed order Transit nodes learn topology by monitoring route records If a transit node already cached the path to the destination, it can directly return it in the RREP as a shortcut d a t t c 7 4 8 b 6 RREQ broadcast at time t RREP via reverse path at time t e
Ad-hoc networking Performance comparison of proactive & reactive protocols Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Proactive SV vs. reactive AOV: 0 nodes 0 sources 4 packets/sec 04 bytes/packet Speed: 0-0 m/sec (Random Waypoint) Pause time: 0 means high node mobility 900 means nodes standing still Broch et al. (998): A Performance Comparison of Multi-Hop Wireless Ad Hoc Network Routing Protocols
Ad-hoc networking Comparison of AOV and SR Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 AOV data throughput SR data throughput AOV control load SR control load AOV SR control overhead > data throughput 00 nodes 40 sources packets/sec bytes/packet Speed: 0-0 m/sec (Random Waypoint) as, Perkins, Royer (000): Performance Comparison of Two On-demand Routing Protocols for Ad Hoc Networks
Vehicular Ad-hoc networking Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009
Vehicular ad-hoc networking Characteristics Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 S Moving nodes with no pre-defined links estinations are not pre-defined, i.e. unknown destination I Path from source to the potential destination does not have to be present the whole time
Vehicular ad-hoc networking Peculiarities of vehicular ad-hoc networks Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Link instability: Links are highly unstable due to movement of nodes. E.g. nodes driving in the opposite direction with a velocity of 0 km/h will have a link for less than 9 s (communication range = ~00m) Limited throughput: Throughput of wireless communication is limited control packet overhead has to be low ata rate (default: MBit/s) has to be shared by all vehicles in the ad-hoc network Error-prone communication: Packet loss due to hidden-terminal problem, interference, shadowing, etc. unreliable communication Recipient selection: Nodes that should receive a packet are a priori unknown (e.g. who should receive an aquaplaning warning)?
Vehicular ad-hoc networking Exemplifying use cases Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Use Case Relevant recipients Relevant timing Cooperative Awareness Traffic Jam Ahead Warning Pre-Crash Sensing Instant Messaging Advertising (e.g. fuel price) All nodes with a similar or overlapping trajectory now or in the near future All nodes approaching the traffic jam in the near future One node that will be involved in the crash One node which is the messaging partner All nodes Short relevance time due to frequent changes Long relevance time due to long lasting event Short relevance time due to criticality Long relevance time Long relevance time due to semistatic information Communication Characteristics One-time best effort Try to keep message in the relevant area for a long time At least feedbacked one-time communication Guaranteed communication with retry Periodic broadcast
Vehicular ad-hoc networking Requirements A routing protocol for vehicular ad-hoc networks needs to be: Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Adaptive: Adaptive to frequent topology changes Fully-distributed: no necessity of centralized controllers Fast: short delays Effective: no message loss, no getting stuck Efficient: avoidance of unnecessary communication Resource saving: Low consumption of storage-space and computation power
Vehicular ad-hoc networking Geographic routing Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Problem: Uninformed search (i.e. a link gives no indication where to find the destination) in flat routing has to use uncoordinated flooding in order to find the destination high bandwidth consumption In contrast to wired networks, network topology in wireless networks is mainly based on geographic position of nodes Idea: Geographical information can be used to direct the search towards the destination Informed search Heuristic is required that maps the position of all next hops to a cost metric in order to make optimal decisions (heuristics try to approximate a globally optimal solution by making a locally optimal decision) Positions can be obtained by a GPS device Nodes periodically broadcast beacons including their geographic position to all nodes in the vicinity (periodicity can be adapted to topology changes)
Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Vehicular ad-hoc networking Geographic routing Shaded area: communication range of source S ashed circle: every node within this circle has a positive progress towards the destination Greedy algorithms Most Forward Within Region Ultimate greedy method Heuristic: Select the node which has the maximum (geogr.) progress to Minimizes the number of hops a packet has to traverse Example: node Nearest Within Forward Progress Heuristic: Select node that is closest to S Minimizes required transmission power reduces collisions Example: node 7 Compass Heuristic: Select nearest node that is directly between S and Minimizes the spatial distance a packet travels Example: node 4 Füßler (007): Position-Based Packet Forwarding for Vehicular Ad-Hoc Networks Mauve et al. (00): A Survey on Position-Based Routing in Mobile Ad-Hoc Networks
Vehicular ad-hoc networking Greedy Perimeter Stateless Routing (GPSR) Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Greedy forwarding decision based on Most Forward Within Region Each node only has to know its one-hop neighbors position (nearly stateless) in order to forward a packet towards its destination scales well with increasing number of destinations Problem: If there is no node within the communication range nearer to the destination (no positive progress to destination), the algorithm will fail (local maximum), although there might be a non-greedy path to the destination Recovery Strategy: Perimeter Routing Route around the perimeter of the void If a next-hop forwarder is found which has a positive progress towards the destination, Greedy Routing is reinstated Perimeter Routing Greedy Routing Füßler (007): Position-Based Packet Forwarding for Vehicular Ad-Hoc Networks
Vehicular ad-hoc networking Greedy Perimeter Coordinator Routing (GPCR) Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 GPSR not suitable for city scenarios Geographic information such as maps (road topology) to select the next forwarder can be used additionally to the neighbors position in the next-hop selection Packets are routed along road segments Nodes at junctions (coordinators) can be used to change a road segment (e.g. node a) C. Lochert, M. Mauve, H. Fussler, and H. Hartenstein (00): Geographic routing in city scenarios
Vehicular ad-hoc networking Location resolution Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 In order to address a specific node, a location resolution mechanism has to be found (similar to route discovery): Location Flooding Grid Location Service (GLS) Homezone Problem: Location resolution (what is the location of node XYZ?) may cause high control packet overhead But often a message (e.g. traffic jam warning, black ice warning) does not need to be addressed to a specific node, but to one/any/all nodes at a specific location area (geo-casting)
Vehicular ad-hoc networking Geo-casting Geographic unicast: Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Geographic broadcast: Geo-anycast + geo-broadcast: Car--Car Communication Consortium (008): Manifesto Addressing of one specific location Addressing of all nodes within a target area Geo-Anycast: Addressing of exactly one arbitrary node within the target area
Summary esign criteria for multi-hop networking Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Route setup: with route setup: proactive: permanent route establishment reactive: route establishment on demand hybrid: combination of proactive and reactive route establishment without route setup (direct forwarding) Addressing: identity-centric: addressing of nodes by their identities geographical: addressing of nodes by their geographic location data-centric: no node addressing, but addressing of data Forwarding decision: source routing: sequence of forwarders is entirely determined by the source node per-hop: next hop is selected by each forwarder based on its local knowledge Address structure: flat: no addressing structure hierarchical: hierarchical structuring of the address space heuristical: addresses allow to make local approximations for global optimum Costs metrics: hops: number of links to traverse delay: delay of packet transmission data rate: data rate provided by links link stability: temporal persistency of a link packet loss rate: quality of a link monetary costs: costs to use a link (e.g. GSM/UMTS)
Summary Routing Protocols for Ad-Hoc Networks Routing Protocols for Ad Hoc Networks Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Table-riven (Proactive) Based on Routing Information Update Mechanism SV WRP CGSR STAR OLSR FSR HSR GSR On-emand (Reactive) SR AOV ABR SSA FORP PLBR Hybrid CEAR ZRP ZHLS Based on the Use of Temporal Information for Routing Path Selection Using Past History SV WRP STAR SR AOV FSR HSR GSR Path Selection Using Prediction FORP RABR LBR Based on Topology Information Organization Flat Routing SR AOV ABR SSA FORP PLBR Hierarchical Routing CGSR FSR HSR Miscellaneous Classification Based on Utilization of Specific Resources Power-Aware Routing PAR Routing Using Geographical Information LAR Routing with Efficient Flooding Table-riven OLSR Murthy & Manoj (004): Ad Hoc Wireless Networks: Architectures and Protocols On-emand PLBR
Questions Lecture Vehicle Networks, Thomas Strang and Matthias Röckl, WS 008/009 Why are classical routing protocols (e.g. RIP, OSPF) not suitable for VANETs? Why V algorithms preferable over LS algorithms for ad-hoc networking? What is the difference between proactive and reactive routing protocols? Why are proactive routing protocols such as SV not suitable for VANETs? Why do geographic routing protocols perform better in VANETs? What are main design criteria for routing protocols?