You are on page 1of 9

Routing in 802.

16 Mesh Networks: A Survey Paper

Jeevan Chalke Avadhoot Punde Kedar Rudre
06329011 06329029 06329036

Department of Computer Science and Engineering

Indian Institute of Technology, Bombay, Mumbai 400076, India
Email: {jeevan, avadhoot, kedarrudre}

Abstract— The IEEE 802.16 WiMax standard provides a The primary focus in such environments is to provide scalable
mechanism for creating multi-hop mesh network, which can routing in the presence of static as well as mobile nodes.
be deployed as a high speed wide area wireless network. The The 802.16 mesh networks are mainly used to provide a cost
network topology is a tree rooted at the base station and the
problem is to determine the routing and link scheduling for the effective Internet access for sparsely populated areas. Thus,
tree, either jointly or separately. The utilization of WiMax mesh the network topology is a tree rooted at the base station and
network can only increase if we efficiently design the multi hop the problem is to determine the routing and link scheduling
routing and scheduling. For effective scheduling first we have to for the tree, either jointly or separately. In community wireless
design the routing policy on top of which scheduling takes place. networks, most of the nodes are either stationary or minimally
The goal of this paper is to present some routing algorithms
proposed by various authors for IEEE 802.16 mesh networks. mobile. Hence the focus of routing algorithms is on improving
In this paper we discuss the routing algorithm for throughput the network capacity or the performance of individual trans-
maximization, for providing QoS (Quality of Service), for fers, instead of coping with mobility or minimizing power
minimizing interference, etc. in detail. usage. One of the main problems facing such networks is the
reduction in total capacity due to interference between multiple
Keywords: IEEE 802.16 Networks, Mesh Networks, Routing
Algorithms, Quality of Service simultaneous transmissions.
There are few fundamental challenges in routing over wire-
less mesh networks. Routing design has to address issues
I. I NTRODUCTION in both short and long time scales. A good wireless mesh
routing algorithm has to ensure both long-term route stability
The IEEE 802.16 protocol, for wireless metropolitan area and achieve short-term opportunistic performance. Wireless
networks (WMAN) has been recently standardized to meet routing also has to ensure robustness against a wide spectrum
the needs of wireless broadband access. The 802.16d [1], also of soft and hard failures, ranging from transient channel out-
known as WiMax (Worldwide Interoperability for Microwave ages, links with intermidiate loss rates, from several channel
Access), supports a point-to-multipoint (PMP) topology and disconnections, nodes under denial-of-service (DOS) attacks,
a mesh topology. It also provides a scalable solution for and failing nodes.
extension of a fiber-optic backbone. WiMax base station can So challenges in routing is to address both these issues and
offer greater wireless coverage of about 5 miles, with LOS at the same time it should be scalable enough to large node
(line of sight) transmission within bandwidth of upto 70Mbps. population. The IEEE 802.16 standard defines the random
WiMax technology can be used for creating wide-area wireless routing, in which SSs randomly select the parent while build-
backhaul network. When a backhaul-based WiMax is deployed ing a tree. In this paper we present some routing algorithms
in mesh mode, it not only increase the wireless coverage but proposed by various authors in their literatures which deal with
also provides features such as lower backhaul deployment some of these issues.
cost, rapid building, easy deployment, robustness and re- The organization of the paper is as follows. In section II,
configurability. This will make it one of the indispensable we present an overview of IEEE 802.16 mesh mode. From
technology in next generation networks. section III to section VIII we present some routing algorithms.
WiMax is sutaible for many neighborhoods that are too We present summary of these routing algorithms in section IX.
remote to receive Internet access via cable or DSL, and for Finally in section X we present our conclusions.
anyplace where the cost of laying or upgrading landlines
to broadband capacity is extremely expensive. In areas with II. OVERVIEW OF 802.16 M ESH M ODE
cable or DSL access, WiMax will provide consumers with an The IEEE 802.16 provides both PMP and mesh technolo-
additional, and possibly cheaper alternative. gies. The main diffrenece between PMP and mesh modes is
Routing in ad-hoc wireless networks has been an active area that in PMP mode, traffic only occurs between the BS (base
of research for many years. Much of the original work in station) and SSs (subscriber stations) where as in mesh mode
the area was motivated by mobile application environments. traffic can be routed through other SSs. As shown in Fig. 1, the
symbols, where MSH-CTRL-LEN in the length of the 802.16
mesh control plane. A scheduled allocation consists of one or
more minislots.


A. Motivation
Multiple-access interference is a major limiting factor for
wireless communication systems. Interference in wireless sys-
tems is one of the most significant factors that limit the
network capacity and scalability. The motivation of the paper
[3] is to design an efficient multi-hop routing and scheduling
scheme that is interference aware, and hence maximizes par-
Fig. 1. A typical mesh network
allel transmission, providing high throughput and scalability.

B. overview
The scheme proposed in [3] includes a novel interference-
overall area is divided into meshes and managed by a single aware route construction algorithm and an enhanced central-
node, which we refer to as Mesh BS (MBS). It serves as the ized mesh scheduling scheme, which consider both traffic
interface for WiMax based mesh to the external network. A load demand and interference conditions. This provides better
transmission can take place between two SSs within a mesh spatial reuse and hence higher spectral efficiency. The scheme
or within two different meshes. The transmission between is based on tree based routing framework. The paper considers
two SSs within a mesh can occur via other SSs within the WiMax-based mesh which is managed by Mesh BS. The
mesh which may or may not involve the MBS. Transmission metric considered for routing is blocking metric B(k). The
between two SSs in two different meshes involves transmission Blocking Metric B(k) of a multihop route indicates the number
from SS to MBS (possibly via other SSs within the mesh), of blocked/interfered nodes by all the intermediate nodes along
from MBS to BS, from BS to MBS of receiver mesh and the route from the root node towards the destination node k.
finally from that MBS to the receiver SS. The paper defines blocking value b(η) of a node η, as the
A. Mesh Mode Frame Structure number of blocked/interfered nodes when η is transmitting.
Thus blocking metric of a route is summation of the blocking
The IEEE 802.16 Mesh mode MAC supports both central-
values of nodes that transmits or forwards packets along the
ized scheduling and distributed scheduling. Most of the algo-
rithms discussed here focus on the centralized mesh scheme
to establish high-speed broadband mesh connections, in which C. Algorithm
MBS coordinates the radio resource allocation within the mesh
The algorithm consists of two parts:
1) Interfence-Aware Route Construction: The scheme se-
Contrary to the basic PMP mode, there are no separate
lects routes with less interference. To do so, the blocking
downlink and uplink subframes in the mesh mode. The mesh
metric for the different routes to the destination from the
mode only supports Time Division Duplex (TDD) to share
source are computed. Then the route with the least value of the
the channel between the uplink and the downlink. A mesh
blocking metric is selected as it will cause least interference
frame consists of a control and a data subframe. The control
as compared to other routes. The example for the computation
subframe serves two functions: network control and schedule
of blocking metric is shown in Fig. 3. In this example the first
control. The data frame is shared between centralized and
path is selected.
distributed scheduling.
Fig. 2 shows the IEEE 802.16 mesh frame structure.
In a network control subframe, mesh network configuration
(MSH-NCFG) and mesh network entry (MSH-NENT) packets
provide some basic level of communication for nodes to
exchange network configuration information. In a schedule
control subframe, the mesh centralized scheduling (MSH-
CSCH) and mesh centralized scheduling configuration (MSH-
CSCF) packets are used for transmission bursts corresponding
to centralized messages, and rest is allocated to transmission
bursts containing mesh distributed scheduling (MSH-DSCH)
packets for distributed scheduling. The data subframe consists
of minislots. Each minislot, except the last minislot, consists Fig. 3. Example of Blocking metric B(k)=2+4+3+4=13; B’(K)=2+4+5+3=14
of d(OFDM symbols/frame − MSH-CTRL-LEN ×7)/256e
Fig. 2. 802.16 Mesh frame structure (Source: [4])

The new node chooses the potential Sponsoring Nodes D. Advantages

based on the blocking metric information. It selects a spon- The algorithm is compared to the basic random scheduling
soring node with least blocking value. described in the IEEE 802.16 standards. They have also com-
2) Interference-Aware Scheduling: The design goal of in- pared it with the theoretical upper bound obtained by linear
terference aware scheduling is to exploit concurrent transmis- programming. In chain topology, the throughput achieved
sion to achieve high system throughput. Let D(k) denotes the outperforms basic scheme and it approaches the upper bound.
capacity request of an SS node from k. D(k) can also be While in case of random mesh topology, the throughput is
represented in terms of Y(j) for every link j. In each allocation better than basic scheme but it is less than the upper bound
iteration t, the scheduling algorithm determines set of active obtained by linear programming. The algorithm leads to better
links. Then the link with highest unallocated traffic demand is spatial reuse and thus higher spectral efficiency.
selected for next allocation of unit traffic. The interfering links
are excluded. The iterative allocation continues until there is E. Limitations
no unallocated capacity request. The algorithm is shown in The given routing scheme in the paper suffers from the
Fig. 4. limitation that they consider the number of blocking nodes
as routing metric. Even if a blocked node does not have any
packet to send, it is considered for calculation. The number of
blocked nodes does not give you the real picture of interference
in the network. The better metric is to consider number of
packets in the blocked nodes.


A. Motivation
In Interference-Aware routing, the authors considered only
a blocking metric of a route. To overcome this limitation, [4]
extended this idea, and also took the number of packets into
account by defining the blocking metric of the node v to be the
number of blocked nodes multiplied by the number of packets
at the node v. Hence the modified blocking metric for a node
v is: B(v) = (Number of nodes blocked by v) × (number
of packets at v). Finally the path with the minimum blocking
metric is selected as P 0 = argmin B(P ). Here we present the
routing scheme presented in [4].

B. Overview
Fig. 4. Interference Aware Scheduling Algorithm
Authors considered problem of routing based on centralized
scheme in which the MBS acts as a centralized scheduler for
the entire network. The authors used following constrained for The problem of finding the BFS tree such that maximum
any transmission: tree is minimized is solved by treating each pair of adjacent
• A node cannot send and receive simultaneously. layers as independent. Consider, adjacent layers Ui and Ui+1 ,
• There may be only one transmitter in the neighborhood minimum degree subgraph that matches each vertex from Ui+1
of a receiver. |Ui+1 |
has a degree between d e and |Ui+1 |. Thus it can be
• There may be only one receiver in the neighborhood of |Ui |
a transmitter. found efficiently using O(log(|Ui+1 |)) invocations of the flow
maximization problem.
The problem can then be stated as follows. Given G = (V, E),
where set V consists of MBS v0 and SSs {va , v2 , }, such D. Advantages
that (vi , vj ) ∈ E if and only if vi and vj are within the
transmission range of each other. SS vi needs to send w(vi ) We can see that the maximum parallelism routing defined
packets to the base station. The objective is to find a feasible above takes both the interference graph and the traffic condi-
routing tree and a schedule1 for the packets such that the tions into account. Also it can change on the basis of change
number of timeslots required is minimized. in traffic conditions even if the interference graph does not
In proposed system model, the routing tree (scheduling tree) change, which is not the case in Interference-Aware routing
is constructed in two conditions. First, when a new node discussed in previous section, where it remains same even if
enters the network, the scheduling tree is updated according traffic conditions change.
to broadcast messages MASH-NCFG and MASH-NENT from
E. Limitations
the new node. Then the MBS recalculates the routing node and
reconfigures the network by broadcasting the MASH-CSCH The disadvantage of these two routing algorithms is that
message to the SSs. Second, the MBS also periodically recom- whenever a new node enters into the mesh, reconfiguration
putes the routing tree by considering new updated throughput of the network takes place which requires broadcasting of
requirements, and changing the routing tree if required. Here control messages (MSH-NCFG and MSH-NENT) resulting in
we present two routing algorithms given in the paper. flooding of network. Algorithms presented here periodically
recomputes the routing tree resulting in extra overhead. Also
C. Algorithm
period at which rerouting should take place is unanswered.
1) Maximum Parallelism Routing: The motivation behind Small period may result in extra overhead on network. Recom-
designing this routing algorithm is to maximize the parallelism puting is based on traffic conditions, thus as traffic changes,
and at the same time taking the number of packets into recomputation of routing tree takes place and hence in the
account. Layered graph can be achieved by constructing the network where traffic is varying or where mobile nodes are
breadth first tree. The authors focused on the set of edges present these algorithms will perform poorly.
between two consecutive layers to find the pairs of edges that
are interfering and pairs of edges that are non-interfering. Each V. ROUTING FOR T HROUGHPUT E NHANCEMENT USING
pair of edge is weighted with the number of packets at the C ONCURRENT T RANSMISSION
sender node. The algorithm selects the set of edges, such that
A. Motivation
considering pair of non-interfering edges in the set, the sum
of weights on the edge is maximized. Suppose, Ui is the set Capacity enhancement in mesh network is quite desirable.
of nodes on the i-th layer, and the set of edges between To increase the capacity of multi-hop system, concurrency
Pn the
two layers is E = {ej : 1 ≤ j ≤ m}, where m = i=1 wi . among the multi-hop transmissions needs to be allowed. In
Authors select the set S ⊆ E, such that [5], authors proposed a simple algorithm for SSs to achieve
X concurrent transmission in both uplink and downlink streams.
f (S) = {w(ej ) + w(ek ) | ej , ek non-interfering} (1)
ej ,ek ∈S B. Overview
is maximized. As explained in section IV, a transmission follows the
Another constrained they enforced is that the set of edges principle that there must be only one transmitter among
uses each vertex from set Ui+1 exactly once, while it can use the neighborhood of receiver and only one receiver among
a vertex from set Ui more than once. This corresponds to the the neighborhood of transmitter. In Fig. 5 curved lines
constraints that a node can have more than one child, but can denote the interference by an active link. Let L(x, y) be the
have only one parent in the routing tree. link from x to y, then the interfered links by L(4, 6) are
2) Min Max Degree BFS Tree: In this routing algorithm, L(6, 4), L(4, 2), L(2, 4), L(5, 2), L(3, 1), L(BS, 1), L(BS, 2)
authors considered a breadth first search (BFS), such that the i.e. when node 4 is transmitting data to node 6, the 7 links
maximum degree of the tree is minimized. Here they tried to mentioned above can’t be active to avoid collision. The
combine the advantage of the shortest path (breadth first tree) number of interfered links by L(x, y) is given by I(x, y), so
having least bottlenecks. I(4, 6) = 7 for example.
1 Here we presented only the routing algorithm, scheduling algorithm can To reduce the interference between links and balance traffic
be found in actual paper load, the structure of routing tree plays a key role. In next
Fig. 6. Construction and adjustment of routing tree

Fig. 5. Link interference in routing tree

D. Advantages
It promotes spatial resource reuse, which increases the
section we will see the proposed algorithm based on interfer- overall end-to-end throughput. Also the algorithm proposed
ence to achieve the concurrency and to improve the network by author is simple to implement.
E. Limitations
C. Algorithm for Constructing Routing Tree Whenever a new SS enters into the network, it might change
Consider, BS-z-y-x is a path in routing tree, and Py (x) the interference value of other SSs and hence affected SSs
is the sum of uplink and downlink interference through the again select their father. This might again change the interfer-
path from node x whose father node is y to BS. So Py (x) is ence value of other nodes and again father selection process
calculated by will take place, which leads to infinite looping. Toggling in
selection of father is also possible. Thus this algorithm will
Py (x) = I(x, y) + I(y, x) + Pz (y) (2)
not result in throughput enhancement if such problems are not
In Fig. 5, P4 (6) = I(4, 6) + I(6, 4) + P2 (4). handled properly.
Initially network has only BS and all the SSs enter the
network one by one. When an SS enters into the network all
its neighbors are eligible to be the father node of the entering A. Motivation
SS. In order to minimize the interference, the entering node Good routing algorithm should have the same route for all
should select a father having least interference. So father node the traffic at a node irrespective of its node of origin and
is whether it is real time or data traffic. Also the route should be
Fx = arg min Pi (x) (3) fixed i.e. it should not vary even when the wireless channels
are time varying. To provide QoS guarantees, we will need
where N eighbor(x) is a set of x’s neighbor nodes. This to reserve resources along the path. This is possible only
ensures the minimum interference along the path. if we do not change the route of a connection unless it is
It may happen that after new entrance, the interference absolutely necessary. Based on these, [2] proposed a fixed
value of other mode change i.e. the entry order impacts the routing algorithm which may work well for both real as well
construction of the routing tree. For this the impacted SSs as data traffic which is discussed next.
select father node once more. Fig. 6 represents the process of
entering and adjustment, where node 5 is entry node. After B. Overview
node 5 entered the network, P2 (4) = 46 and P5 (4) = 30, so Let rk (i, j) be the assigned transmission rate, Xk (i, j) the
the father node of 4 is adjusted from 2 to 5. external arrivals and Yk (i, j) the arrivals from other nodes to
node i for output link (i, j) during the frame k. If the schedule
Concurrent Transmission Algorithm
is fixed then the link (i, j) is always assigned ni,j slots in a
Achieving spatial reuse with concurrency is an effectual frame, such that
method to improve the throughput in multi-hop system. After XM

analyzing the construction of routing tree and scheduling ni = N (4)

defined in WiMax mesh network, authors proposed a con- i=1
current transmission algorithm1 with no collision to improve where N is the number of slots in a mesh frame. Also let
the overall end-to-end throughput on top of this constructed λi,j = E[Xk (i, j)]. Then
routing tree.
Qk+1 (i, j) = (Qk (i, j) + Yk (i, j) − ni,j rk (i, j))+ + Xk (i, j)
1 The description and actual algorithm is available in [5] (5)
where (x)+ denote max(0, x) and Qk (i, j) be the queue E. Limitations
length at node i for the output link (i, j) in the beginning
of frame k. For the queue to be stable, following condition Routing algorithm presented above assumes that the route
should be satisfied, is fixed. Thus, if the link failure occurs or route degrades
considerably, which is reasonably high in wireless medium,
ni,j E[r(i, j)] E[X(i, j) + Y (i, j)] = λi,j + E[Y (i, j)] (6) the routing will fail. Authors do not mention any mechanism
to handle such link failure. Recomputing the routing tree may
where the expectation E[Y (i, j)] is under the stationary dis-
solve this issue.
C. Algorithm VII. ROMER
For fixed routing, splitting traffic from a node along two
paths to MBS is not optimal unless both the routes have same A. Motivation
link cost and even then giving the whole traffic to one path
will not make it worse because the channel schedule can be The paper [7] has discussed ROMER, a resilient opportunis-
adjusted accordingly. It will also lead to resequencing delays tic mesh routing for wireless mesh network. The motivation
at the receiver. Thus authors used a single output link which of the paper is to design a routing protocol for wireless mesh
imply that they have a tree structure. Then corresponding to networks, which will meet the following two goals: (1) Re-
node i there will be a unique output link. Thus single index i silience against lossy links, transient, permanent link outages
indicates the output link (i, j). Let and occasionally high-speed Access Point(HAP) failure; (2)
High data rate along the route that exploits receiver diversity
X and the available multi-rate capability at the physical layer.
λi = λi,j (7)
B. Overview
Then mi
E[Yi ] = λai,j (8) ROMER balances between long-term route stability and
j=1 short term opportunistic performance. ROMER exploits the
dense deployment of HAP nodes and builds a randomized
where {ai,1 , ai,2 ,,mi } are the nodes whose data passes and opportunistic forwarding mesh to enhance throughput and
through node i. Thus, if robustness. It is a two-tier architecture for routing that balances
M Pmi !
between long-term optimality and short-term opportunistic
X λi + j=1 λai,j
<1 (9) gain. The mesh is centered around the long-term stable,
i=1 minimum cost path, but expands or shrinks opportunistically at
is satisfied, then the fixed allocation scheme that can stabilize the runtime to exploit high quality, high data rate links enabled
the system is found. rearranging the terms we get by the physical layer multirate options. The high rate link
M hi
! out of the candidate routes are selected for the forwarding of
X X 1 the data. Also redundant data is transferred from other routes
λi . <N (10)
i=1 j=1
E[r(pi,j )] randomly, to ensure resiliency against lossy links and transient
node outages.
where {pi,1 , pi,2 , ...pi,hi } are the nodes in the path through
which data of node i is routed. To maximize the overall
stability region, choose the route that minimizes. C. Algorithm
X 1 The algorithm has two main components:
(11) 1) Building Candidate Forwarding Mesh on the Fly: Each
E[r(pi,j )]
packet based on credit approach builds its forwarding mesh
for each node i. Also, choosing this route reduces the average on the fly. The packet may follow a subset of the candidate
transmission time needed to transmit a packet from node to interleaved paths offered by the mesh to resist against channel
MBS. This routing is fixed for all frames for each node along and node outages. The example of credit based mesh is
the path that minimizes equation (11). This route can be found shown in the Fig. 7. The packet starts from the source node
by standard shortest path routing algorithms such as Dijkstra’s with a fixed credit assigned to it. As it travels through the
or Bellman-Ford by assigning cost E[r(p1 i,j )] to link (i, j). different links of the mesh, the credits get subtracted. The
value subtracted depends upon the credit assigned to the links
D. Advantages and the nodes. In the credit scheme discussed in the paper,
The algorithm presented has a shortest path between SSs more credits are consumed in the initial hops and less credits
and MBS and this path remains fix for all the traffic. Due to to the hops which are near the destination. This narrows down
fixed path, reservation of resources is possible and hence QoS the candidate mesh. Example of mesh based forwarding is
guarantees can easily be provided. shown in the Fig. 8.
solutions that are proposed in the literature to overcome this
problem. Comparison with these algorithms would have given
better insight of the efficiency of the protocol. In ROMER
packets travels from different paths, so it involves overhead
and delay of resequencing of packets at the destination. Hence
this protocol is not suitable for real-time traffic. Also the
paper does not make any comments or comparison regarding
the delay incurred by packets which is an important QoS
parameter. The protocol acheives reliability by transmitting
redundant data. The simulation topology that they have con-
sidered consists of only one source and destination pair. Also
they consider low load scenario. It will be interesting to see
Fig. 7. Credit-based runtime mesh the performance of the algorithm in case of multiple source
destination pair and high load scenario. In such condition
transmission of redundant packets may lead to wastage of BW,
more delay and less throughput.
The credit based approach used in the algorithm requires
to store credit for every link for each source destination pair.
Thus with large number of source and destination pair, the
overhead of maintaining credits is going to be large. These
limitations clearly suggest that the algorithm is not scalable.
Also the policy for assigning credit is not according to the state
of the network. It will be better to assign credits according to
the interference pattern. This will also remove the overhead
of storing the credits for each source-destination pair. Also
while comparing with the multi-path routing, the algorithm
considered is 2-disjoint-path routing. A more sophisticated
algorithm would have given a better picture.
Fig. 8. Credit-based runtime mesh
2) Randomized Opportunistic Forwarding: The goal of A. Motivation
opportunistic forwarding is to leverage the short term channel Community wireless networks are characterized by station-
diversity to select the highest throughput link on the runtime ary nodes which do not rely on batteries. In these envi-
mesh. Once the mesh is constructed for a data packet using ronments, the focus of routing algorithms is on maximizing
credit mechanism, each intermediate node selects the best throughput rather than coping with mobility or minimizing
downstream link, which offers highest instantenous rate and energy. Traditionally, shortest path algorithms are used to
forwards the packet over this link with probability 1. For other provide high throughput. However, they do not perform well
downstream links it calculates the forwarding probability p, when nodes have multi-radio network interface cards. With
and sends the packet over these links with this probability. rapidly diminishing prices of 802.11 network cards, it is not
For a link l the forwarding probability l is calculated as surprising to have multi-radio network interface cards on a
Rl single machine. Thus new routing algorithms are necessary to
pl = (12) increase throughput.
Prior research by De Couto uses Expected Transmission
D. Advantages Time (ETT) metric to provide high throughput. It does perform
Romer creates runtime mesh for a packet and also pro- better than shortest path metric in single-radio environment.
vides higher throughput by opportunistically selecting the However, it does not perform well in multi-radio environment.
instantaneously higher quality wireless link. ROMER can Considering all these factors, a new protocol is proposed
achieve about 68-195% higher throughput gain over single- called Multi-Radio Link-Quality Source Routing (MR-LQSR)
path routing. It is more robust than single and multi-path protocol.
routing protocols. It provides better packet delivery ratio as
compared to single-path routing and multi-path routing. B. Overview of Routing Protocol
1) Design Goals: Firstly, the protocol should consider the
E. Limitations and extensions loss rate and bandwidth of the link. The transmission time
The single path routing with which they have compared of packet depends on both, bandwidth of the link and loss
suffers from the HOL blocking problem. There are several rate. Additionally, path metric should be increasing. It means,
if a hop is added to the existing route, the cost of the path Calculation of WCETT: Here we are combining individual
should never decrease. Thirdly, it should consider reduction in ETTs into a single metric that should reflect overall goodness
throughput due to interference between links that operate on of the path. Considering design goals,
the same channel. n
The functions of protocol can be summarized into four W CET T = ET Ti (17)
components viz. i=1
1) A component that discovers the neighbors of the node. Here we are not considering hops on different channels. We
2) A component that assigns weights to the links a node can assume that two paths on same channel always interfere.
has with its neighbors. Therefore, considering channel diversity, we get
3) A component to propagate this information to other
nodes. W CET T = max Xj (18)
4) A component that uses link weights to find a good path
for a given distribution. Where
The first and third components are similar to DSR. Second X
Xj = ET Ti 1<j<k (19)
component uses Expected Transmission Count (ETC) to assign
Hop i is on channel j
link weights. Fourth component uses Weighted Cumulative
Expected Transmission Time (WCETT). These two equations can be combined to give
2) The path metric: n
Calculation of ETX: Let us denote packet loss probability W CET T = (1 − β) ∗ ET Ti + β ∗ max Xj (20)
in forward direction by pf and in reverse direction by pr. If i=1
p denotes the probability of unsuccessful packet transmission
where, 0 ≤ β ≤ 1
from x to y,
This equation can be seen as a tradeoff between throughput
p = 1(1 − pr) ∗ (1 − pf ) (13) and delay. First term is sum of transmission times along the
hops. This reflects latency along the path. The second term
Let probability that packet will be successfully delivered from represents bottleneck hops and therefore it impacts throughput
x to y after k attempts be denoted by s(k) of the system. The weighted average is an attempt to get
balance between the two.
S(k) = pk−1 ∗ (1 − p) (14)
C. Advantages
Thus, expected number of transmissions required to success-
fully deliver a packet from x to y is The protocol is a compromise between throughput and
delay. The equation consists of two terms. First term in-
∞ dicates delay factor and second term indicates throughput.
X 1
ET X = k ∗ s(k) = (15) The value of β indicates weightage given to throughput or
i=1 delay. Traditional protocols use per neighbor probing which
has complexity O(n2 ). However, this protocol uses broadcast
This ETX is multiplied by link bandwidth to get Expected
probing which is O(n).
Transmission Time (ETT).

S D. Limitations
ET T = ET X ∗ (16)
B The derivation of ETT does not consider impact of con-
tention. The contending traffic may increase packet loss rate
where, S = size of packet and B = bandwidth
due to collisions. It also reduces available bandwidth. The
To calculate ETT, a node needs to know values of pf
derivation also considers packet loss rate as an independent
and pr. These values are computed using broadcast packet
parameter whereas in reality, it may depend on channel
technique. Each node periodically sends out broadcast node
utilization. It also uses total link capacity instead of available
packet. Nodes keep track of number of broadcast packets
received from their neighbors and add this information in
their packets. Thus they come to know number of packets
successfully transmitted and calculate pf and pr. To measure
bandwidth, the technique of packet pairs is used. Each node We presented various routing algorithms for IEEE 802.16
sends two back to back probe packets to its neighbors. The first mesh networks. In that we presented the motivation behind
packet is small whereas second packet is large. The neighbor the designing of the routing algorithms, then overview of the
measures time difference between reception of two packets routing algorithms and finally some limitations associated with
and sends the value back to the sender. Few samples of these them. In table I, we have presented the comparison/summary
values are taken and bandwidth is estimated. of these algorithms.

Algorithm Key Idea Advantages Limitations

Interference-Aware Performs routing based Higher throughput, spatial reuse The metric considered does not give
Routing [3] on the knowledge of in- and higher spectral efficiency the complete picture of the interfer-
terference ence
Routing for To maximize the Considers the traffic characteristics Several tree reconfigurations lead to
Throughput throughput by i.e. number of packets at each node extra overhead
Maximization maximizing parallelism along with the blocking metric
Routing for Minimizing link interfer- Concurrent transmission promotes The process of entering and adjust-
Throughput ence spatial reuse, which increases the ing may lead to infinite looping and
Enhancement end-to-end throughput toggling between two fathers
using Concurrent
Transmission [5]
Fixed Routing for To provide QoS Route is fixed for both real and Link failure is not handled, recon-
Supporting QoS data traffic, shortest path between figuration of routing tree is required
[2] SS and MBS
ROMER [7] Achieves long-term Better robustness Resequencing overhead, redundant
route stability and packets, not scalable
short-term opportunistic
Multi-Radio Link- Link cost is function of Compromise between local Impact of contention is not consid-
Quality Source bandwidth and loss rate optimization (shortest path) ered, packet loss rate is considered
Routing (MR- and global optimization (avoid as an independent parameter
LQSR) [8] bottleneck hops), Uses broadcast
probing that is of order O(n)

[1] IEEE Standard 802.16-2004, “IEEE Standard for Local and Metropolitan
Area Networks - Part 16: Air Interface for Fixed Broadband Wireless
Access Systems”, Tech. Rep., May 2004.
In this paper we have surveyed six routing protocols for [2] H. Shetiya and V. Sharma, “Algorithms for Routing and Centralized
WiMax based Mesh networks. Routing in WiMax is an active Scheduling to Provide QoS in IEEE 802.16 Mesh Networks,” in
area of research. Many schemes have been proposed which WMuNeP ’05: Proceedings of the 1st ACM workshop on Wireless
multimedia networking and performance modeling, New York, NY,
try to increase throughput, minimize delay and provide more USA, 2005, pp. 140-149, ACM Press.
robustness over wireless channel. [3] H. Wei, S. Ganguly, A. Izmailov, and Z. Haas, “Interference-Aware IEEE
802.16 WiMax Mesh Networks,” in Vehicular Technology Conference,
In papers [3], [4], and [5], the routing algorithms try to 2005. VTC 2005-Spring. 2005 IEEE 61st, Vol. 5; 3102-3106.
achieve higher throughput by making use of better spatial [4] F. Jin, A. Arora, J. Hwang, and H.-A. Choi, “Routing and Packet
reuse and higher spectral efficiency. Such algorithms are better Scheduling for Throughput Maximization in IEEE 802.16 Mesh Net-
works,” submitted for publication.
in dense network where throughput of the system is less [5] J. Tao, F. Liu, Z. Zeng, and Z. Lin, “Throughput Enhancement in
due to high interference. ROMER [7] provides an adaptive WiMax Mesh Networks Using Concurrent Transmission,” in Interna-
algorithm that gives better robustness against lossy wireless tional Conference on Wireless Communications, Networking and Mobile
Computing, Sep 2005, pp. 871-874.
links by transmitting redundant packets in controlled fashion. [6] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless Mesh Network: A
This algorithm will work well where wireless links are lossy. Survey,” January 2005..
But ROMER has overhead of resequencing which results into [7] Y. Yuan, H. Yang, S. Wong, S. Lu, W. Arbaugh, “ROMER: Resilient Op-
portunistic Mesh Routing for Wireless Mesh Networks,” in Proceeding
more delay and hence cannot be used to support real time of IEEE Workshop on Wireless Mesh Networks (WiMesh), 2005.
applications. In paper [8] the algorithm tries to obtain trade- [8] R. Draves, J. Padhye, and B. Zill, “Routing in Multi-radio, Multi-
off between throughput and delay by considering metric based hop Wireless Mesh Networks,” in ACM MobiCom, Philadelphia, PA,
September 2004.
on shortest path as well as bottleneck hops. Such algorithms
can be used to support both data as well as real time traffic
by adjusting the tunable parameter. The paper [2] provides
fixed routing which enables resource reservation for providing
QoS, hence the protocol suggested in the paper can be used
to support real time applications.