Wireless Mesh Networks Anatolij Zubow (zubow@informatik.hu-berlin.de) Introduction to Wireless Mesh Networks

Problem Statement Path Selection Medium Access Control Generic routing algorithms form the basis of most routing protocols (Bellman, Dijkstra, Floyd) Metrics used by these protocols to calculate the optimum route: hop count, link speed, delay, … Wireless networks Wireless links are unstable Topology (connectivity) and link speed change quickly (fading + interference) Routing graphs have short periods of availability Known path metrics: packet error rate (ETX), expected transmission time (ETT), availability of relay node on certain frequency channel Medium Access Control Multi-hop networks Frame exchange among devices that are not in mutual Rx range 2

Capacity of the Radio Channel Spatial Reuse Frame directed by a node to a non-direct neighbor must be relayed (increasing spectrum load + overhead) Each relay node operates for each relayed frame as both receiver and transmitter Wireless medium around relay node is occupied once for frame reception and a second for frame transmission 3

Spatial Reuse Assuming FS-PL, the interference range may be assumed to be less twice the RX range With constant transmission duration and equally spaced devices an optimal spatial frequency reuse distance can be defined Since actual SNIR value determines success or failure of a frame reception, interference prediction is crucial to be able to exploit spatial frequency reuse 4

Hidden Terminal Problem A node in the mesh network can communicate to only a subset of all devices However, it can interfere many more others Hidden station (STA): An STA whose transmissions cannot be detected using Carrier Sense (CS) by a second STA, but whose transmissions interfere with transmissions from the second STA to a third STA 5

Hidden Terminal Problem The receiving device is in reception range of two other devices However, the latter ones are out of mutual reception range Hence, transmissions to the RX device cannot be detected by a possible interferer Solution: channel reservation (RTS/CTS) 6

Busy Tone Concept Dedicated signaling channel Use an additional narrowband transceiver to transmit a busy tone with long reach to avoid any interference from hidden devices Whenever a device is receiving, it sends a non-modulated busy tone in the narrow band channel All devices must listen to the narrowband channel prior to any transmission attempt If the busy tone is sensed, no transmission is allowed B transmits a busy tone on a different channel. Device C, which is outside reception of the transmitting device A, receives the busy tone and defers from channel access. 7

Exposed Terminal Problem A device is called exposed if according to the protocol applied, the device decides that the channel is not available, although its transmission to another ongoing transmission would not cause harmful interference Reduces spatial reuse by prohibiting parallel transmissions (medium is wasted) 8

Fairness Fairness denotes a specific means of resource sharing Different characteristics of fairness may be distinguished: Based on frames (independent of size, PHY mode) all frames have equal chance to get access to the medium (e.g. 802.11) Based on capacity (transmission duration) all devices have equal share in terms of transmission duration (e.g. 802.11e TXOP) Based on priorities Based on flows?!? 9

Congestion When the number of flows is large and the capacity is fixed, none of the traffic flows may be able to fulfill its QoS Flow Admission Control (FAC) is required: Denies access of additional flows to the network Unfair to newly arriving flows Traffic flow prioritization (starvation of lower prioritized flows) Sometimes the processing speed of devices might limit the performance 10

Transmission Control Protocol (TCP) TCP greatly impacts the performance of wireless networks TCP was designed for wired networks: Congestion detection + avoidance algorithm is based on the assumption of frame losses in congested routers (insufficient capacity) Frame error rate is assumed to be low and constant Wireless networks are different: Fluctuations in channel condition + mobility result in high variance of frame error ratio Link adaption changes PHY modes and thus the link capacity TCP draws the wrong conclusion and reduces the windows size making wireless links appear weak in capacity TCP and ARQ (802.11) are working against each other (interference) Congestion avoidance in packet-based wireless mesh networks 11

Routing Wireless networks operating at high frequencies suffer from small coverage areas Only nodes in close vicinity are able to communicate directly Otherwise multi-hop forwarding is required Routing Algorithms: Proactive protocols constantly maintain and detect paths to all possible destinations Reactive protocols save overhead and set-up paths only when needed Hybrid protocols combine both aspects (“Zone Routing Protocol”) Further classification: Link state and distance vector based algorithms, Hierarchy of path selection entities, Usage of location information for forwarding decisions 12

Routing (2) Proactive protocols: high overhead, low latency Reactive protocols: low overhead, high latency Route breakage When a route in use breaks, both proactive and reactive protocols need to recover by establishing a new route Most approaches inform the source node and it then starts a completely new route discovery process (network is flooded with large number of signal messages) Other protocols try to keep the route discovery local around the node involved All approaches only react in a proper manner, when the route is already broken (high number of lost packets) 13

Dynamic Source Routing When A sends a data packet to F, the entire route is included in the packet header Intermediate nodes use the source route embedded in the packet’s header to determine to whom the packet should be forwarded Different packets may have different routes, even they have the same source and destination → Hence called as dynamic source routing 14

DSR Basics Two basic mechanisms Key optimization Route Discovery Route Request (RREQ) Route Reply (RREP) Route Maintenance Route Error (RERR) Key optimization Each node maintains a route cache Overhears data, RREQ, RREP, and RERR packets Passively collects new routes as many as possible Reduces the cost of Route Discovery and Route Maintenance 15

Route Discovery When to perform a Route Discovery ? Every route request packet (RREQ) contains <target address, initiator address, route record, request ID> Each node maintains a list of the < initiator address, request ID> When a node Y receives a RREQ Discards the route request packet if < initiator address, request ID>is in its list Returns a route reply packet which contains a route from initiator to target If Y is target If Y has an entry in its route cache for a route to target Appends itself address to the route record in RREQ and re-broadcast RREQ 16

Route Discovery in DSR 17

Route Discovery in DSR 18

Route Discovery in DSR 19

Route Reply in DSR 20

More Details For Route Reply in DSR Destination F on receiving the first RREQ, sends a Route Reply (RREP) RREQ includes the route from A to F How Route Reply packet is sent to A? Route Reply can be sent by reversing the route in Route Request (RREQ) If links are bi-directional If unidirectional(asymmetric) links are allowed, then a route to A is needed Local route cache has a route to A Piggybacking Route Reply in Route Request packet for A NOTE: If IEEE 802.11 MAC is used, then links have to be bi-directional 21

Route Maintenance 22

More Details For Route Reply in DSR Hop-by-hop maintenance (MAC or network layer) How to find link [node-i,node(i+1)] is down ? Utilize MAC level acknowledgement Passive acknowledge (overhearing node(i+1) re-transmission) How to send route error packet to A? Use the reverse route [node-i,node(i-1), ……,node-1, A] Use node-i route cache to get a route to A Piggybacking route error packet in route discovery packet A 23

DSR Optimizations (1) Route Caching Advantages Problems Each node caches a new route it learns by any means When A finds route[A,B,D,E,F] to F, A also learns route [A,B,D,E] to E, … When G receives Route Request [A,B,D,E] destined for some node F, G learns route [G,E,D,B,A] to A if links are bi-directional D forwards Route Reply [A,B,D,E,F], D learns route [D,E,F] to F When B forwards Data [A,B,D,E,F] it learns route [B,D,E,F] to F Advantages Can speed up Route Discovery Can reduce Propagation of Route Requests Problems With time passing and node moving, cached routes may become invalid (stale caches can adversely affect on network performance) 24

DSR Optimizations (2) Avoid Route Reply Storm problem: When intermediate nodes reply Route Request by using local route cache Pick a random number d=H*(h-1+r) , H= average per hop delay; h= length of the route; r = random number in [0,1] Delay transmitting the route reply for a period of d During this period, cancel the route reply if overhearing a packet contains a route from the same initiator to the same target with length no longer than h 25

DSR Optimizations (3) Expanding Ring (limits the propagation of RREQ packets) First, set TTL = 1 for first route request packet If no route reply is received after some time period, set TTL = maximum for next route request packet 26

DSR Optimizations (4) Reflecting shorter route D may find that route may be shortened by excluding the intermediate hop C, D then sends an unsolicited route reply packet to A to inform the shorter route 27

Ad-hoc On-demand Distance Vector Routing (AODV) 28