Table-driven geographical routing using aggregate

areas
Mandimby Ranaivo Rakotondravelona∗ , Bruno Baynat† , Fanilo Harivelo∗ and Pascal Anelli∗
∗ Laboratoire
d’Informatique et de Mathématiques (LIM)
Université de la Réunion
Email: {mandimby.ranaivo, fanilo.harivelo, pascal.anelli}@univ-reunion.fr
† Laboratoire d’Informatique de Paris 6 (LIP6)

Email: [email protected]

Abstract—Geographical routing protocols scale well in large which has not any neighbor geographically closer to the
ad-hoc and sensor networks but fall short with some topologies. destination, hence forwarding can make no more progress.
This causes packet losses and a drop in network performance. Several techniques allows to circumvent these dead-ends as
In this paper, we propose to introduce topological information
aiming to improve the quality of routing decision while keeping described in [3], among them, flooding, geometric-based or
the protocol scalable. Every node maintains precise information planarization. Nonetheless, they have limitations such as the
for nearby nodes and aggregated information for farther nodes need of bidirectional links or nodes with equal radio range [4].
by means of aggregate areas. Evaluation demonstrates the scal- We investigate the possibility to enrich a position-based
ability of the proposition and its efficiency compared to a pure routing with topological information. With an exclusive
geographical routing protocol.
Index Terms—Ad hoc networks, geographical routing, table-
topology-based approach, each node has a more or less com-
driven routing, aggregation. plete knowledge of every other nodes in the network [5]. More
precisely, a list of existing links in the network is maintained
I. I NTRODUCTION at any time (proactive protocols) or determined when needed
(reactive protocols). Forwarding decisions are taken based
Ad-hoc and sensor networks involve nodes interacting with on these link information. One of the advantages of these
each other over multihop connections without central manage- approaches is the precision as routing is performed only on
ment in an infrastructure-less environment [1]. One of the main known links. But they face scaling problems. As a matter of
issues in such self-organized networks is scalability in terms fact, in large networks, the list of existing links can be huge
of routing. With the numerous potential applications of ad- and too costly to maintain.
hoc and sensor networks (Internet of Things, disaster recovery, On one hand, we aim to improve the quality of routing
military application ...) scalability is an important property. It decisions. On the other hand, we would like to keep the
allows the network to grow in scale without quick performance scalability of the overall protocol. We provide routing with
loss. For instance, a scalable routing protocol still ensures high information about links, and, thus, existing nodes. But not all
packet delivery rate even with increased number of nodes and information is necessary for a given node. Far destinations
also without excessive cost. don’t require high precision routes to be reached as routing
Geographical routing approaches are known to scale well decision is taken on every hop. Let’s imagine a traveler who
in large ad-hoc and sensor networks. They require nodes to wants to join his booked hotel room in a foreign country.
maintain existing links to their one-hop neighboring nodes. But First, he has to travel to the destination country. He doesn’t
additionally, geographical information are needed to perform need very detailed information so far. On arrival, he must find
forwarding. The nodes have to be aware of their location by the town where his hotel is located by consulting, for instance,
means of GPS or alternative techniques [2]. Forwarding can be a map of the country. As he gets closer, he now needs more
summed up as selecting one of the one-hop neighbors which is detailed information. And so on, arrived in the town, he will
geographically closer to the destination. Routing decision are need very sharp details in order to find the quarter and the
purely local. Because the list of existing links to be maintained street where his hotel is located. By analogy, routing needs
by each node is small and network size-independent, these only more and more details as the destination comes closer.
approaches are scalable. However, due to lack and/or limited Based on this principle, we propose a scalable approach as
use of topological information, geographical routing performs it gives more precision to localized routing information while
poorly with some network topologies. Nodes ignore or are summarizing the global ones. For a node, the list of nearby
not aware of the availability of some links in the network. nodes is maintained as reachable hosts in the routing table.
Choosen paths are not necessarily optimal. In extreme cases, Farther nodes are gradually aggregated into larger and larger
forwarding fails even if routes exist to the destination. One areas as they are located away from the considered node. We
noticeable consequence of such failure is the appearance of aggregate nodes by their geographical coordinates.
dead-ends called local maxima. A local maximum is a node Our proposition relies on the following components:

978-1-5386-3831-6/17/$31.00 ©2017 IEEE

 x
x
x 1 2
x 1 1
x 1 1
x 2 2
x
x
x 1 1
x 2
1 2
x 1
x x 1 2
x 1

x 1
x x 1
x
1 2
x 2
x
N x 1
1 1
1 2
2
x 1 1
x
x x
x x 1
x
x 1 2
1 1
1
2
x 2 2
x x
x
x 1 1
1 1
x 1 2
x x
x
x

x (a) Order-1 (b) Order-2

Fig. 2. Example of aggregate areas
Fig. 1. Local area

A node retains link information to all the nodes inside its

• At some point, we need to aggregate some nodes into local area. On the Fig.1, the local area of the node N is colored
area. Aggregation is done on location-basis. We, then, in grey.
partition the whole deployment space into geographical B. Aggregate areas and network zone partitioning
areas, that we call primary aggregate partitions or areas,
Link information to nodes inside the network are aggregated
shared and known by all nodes in the network. Primary
by means of geographical areas called aggregate areas. We use
partitions can be aggregated, in their turn, to form larger
the following principles to design aggregate areas:
partitions to make it possible to refer to faraway nodes.
• We define primary aggregate areas that can be uniquely
They are uniquely identified and known by all nodes in
identified;
the network, and so on;
• Aggregate areas can be, in turn, aggregated into larger
• Known nodes and areas are maintained in a routing table
aggregate areas that are also uniquely identified;
by every node;
• The aggregation level of each area is known.
• Nodes exchange periodically routing information to keep
their routing table up-to-date; We define multilevel aggregate areas: order-n areas are ag-
• An unified forwarding strategy consists in sending a
gregated into order-n + 1 where n is the aggregation level.
packet to the gateway of the entry (either a node or an The smallest aggregate areas (primary or order-1) result from
area) closest to the destination. a global partition of the network zone.
On Fig.2a, the network zone has been partitioned into order-
Mixing topology-based routing and location-based routing
1 aggregate areas. Then on Fig.2b, order-1 areas have been,
was already proposed in the literature [6], [7]. We make differ-
in turn, aggregated into order-2 ones. The result is a global
entiation by using only one type of routing key (the destination
partition of the network into order-2 aggregate areas.
location), a tight integration of topology-based and location-
based components (forwarding decision relies solely on the C. Table structure
routing table) and more flexible and user-defined settings to From a node point of view, the routing table reflects its
tune the tradeoff between precision and cost related to the knowledge of, on one hand, the link information to nodes
maintenance of routing information thanks to aggregation. We inside its local area and, on another hand, aggregated link
demonstrate through simulations that our proposition makes information to nodes inside aggregate areas. Thus there are
larger use of existing paths while keeping the routing table two parts inside the routing table:
size low. • Local area part: containing route to nodes inside the local
area. Each line in this part indicates how to join a specific
II. TABLE - DRIVEN GEOGRAPHICAL ROUTING
node inside this area;
A. Local area • Aggregate areas part: containing information about aggre-

Each node has its own local area. It is a limited geographical gate areas known to contain at least one node. Each line
area surrounding a node. Its size can be defined in terms of hop of this part gives information on how to join a specific
counts, i.e., it is the area containing nodes at a given maximum order-n aggregate area.
number of hops from the considered node. The size can also The Table I sums up the structure of a routing table.
be limited by geographical distances. In this case, the local • Area (A) indicates the nature of the route, either leading
area is defined by d(N, P ) ≤ l where d is a distance function, to a node (local) or an order-n aggregate area (order-n
N is the considered node, P is a point in the network zone aggr.). N is the highest order;
and l is the maximum allowed distance from N to be part of • Destination (D) tells the ID of the destination of the route
the local area. which can be a node or an aggregate area;
 TABLE I of aggregation. More precisely, from a node perspective, the
ROUTING TABLE STRUCTURE further is the node corresponding to a link information, the
Area(A) Destination(D) Gateway(GW) Cost(C) higher is the order of the aggregate area containing it. The
Local Node ID Node ID ... order of the aggregate area is function of the distance between
... the node receiving the link information and the node concerned
Local Node ID Node ID ...
Order-1 aggr. Area ID Node ID ... by the latter:
... 
1 if 0 < d(N, P ) ≤ x1
Order-1 aggr. Area ID Node ID ... 
Order-2 aggr. Area ID Node ID ... 
2 if x < d(N, P ) ≤ x
... 1 2
Order-2 aggr. Area ID Node ID ...
order =
...
... 

n if xn−1 < d(N, P ) ≤ xn

...
Order-N aggr. Area ID Node ID ...
... where N is the center node (owner of the table), P is a point
Order-N aggr. Area ID Node ID ...
in the network zone, d is a distance function and
x1 < x2 < ... < xn are reference distances.
3 The Fig.3 shows the aggregate areas (with their respective
3
order) considered by the node N.
Initialization. At start up, a node adds its own information
2
2
to its table: the area column is filled with local, the destination
with its own ID, the gateway also with its own ID and the cost
1 with zero.
2 1 2 Receiving updates. Let N a node receiving a routing
N
message for a destination D from one of its neighbors noted
3
1 GW. The routing table of N will be referred as T(N). N will
2 2 follow the next algorithm in order to update its routing table
2
given this new information. In the following, C0 is an update
of the cost C after routing update processing.
• If D is a node:
Fig. 3. Aggregate areas of N
– If D is located inside the local area:
∗ If there is not yet route to D inside T(N), add the
• Gateway (GW) is the ID of the next hop node through route {local, D, GW, C0 };
which the destination can be reached; ∗ If there is already a route to D inside T(N), it will
• Cost (C) indicates the cost of going from the current node be replaced if the new route offers a lower cost.
to the destination according to a specific metric (e.g. hop
– If D is located outside the local area:
count).
It is important to note that these information are the minimal ∗ Identify the order-1 aggregate area containing D;
necessary to be retained in case of a static network. Additional ∗ Restart the algorithm with the latter area as entry
information are required to handle mobility (which is out of D.
the scope of this paper) such as sequence numbers [8]. • If D is an aggregate area:

– Find the smallest aggregated area known by N con-

D. Updating tables
taining D, let A the resulting aggregate area whose
Because of the table-based nature of our protocol, one of order is n:
the key points is the updating process of the table content.
∗ If there is not yet route to A inside T(N), add the
We propose an updating mechanism which is similar to what
route {Order-n aggr., A, GW, C0 };
is used in DSDV [8]. Every node periodically broadcasts its
∗ If there is already a route to A inside T(N), it will
table content to all of its one-hop neighbors. Restraining the
be replaced if the new route offers a lower cost.
broadcast to the one-hop neighbors saves bandwidth yet allow
the routing information to spread across the network. Example. On the Fig.4 we have a representation of the
Aggregation decisions. As stated earlier, a node’s routing local area and aggregate areas of a node N. We assume that
table reflects its knowledge of, one one hand, the link infor- N’s table has just been initialized so contains only one entry
mation to nodes inside its local area and, on another hand, (its own). Now, N receives the following routing messages
aggregated link information to nodes inside aggregate areas. In {Destination,C} from a node that will be referred as
other terms, information outside the local area are aggregated. GW:
In order to optimize the scalability of our solution, we pose • {A, C}: A is situated inside the local area, so N adds
the following principle: with great distance comes high level the line {local, A, GW, C0 } to its table;
 3 3

3 D 3
area_2
2 2
2 area_gw 2

B area_2
area_1
1 1
A area_1
2 1 2
GW 2 A 2
1

N N
3 3
1 1
B
2 2 2 2

2 2

Fig. 4. Receiving routing information Fig. 5. Example of forwarding

• {B, C}: B is outside the local area but inside gateways are at further distances from the destination than
area_1. Thus, the routing table is updated with the source. But the failure rate is expected to be lower. There
{Order-1 aggr., area_1, GW, C0 }; are two main reasons for that. First, the reachable areas are
• {area_gw, C}: area_gw is an order-1 aggregate not restrained to the immediate neighborhood of a node. They
area. At this distance, N doesn’t retain any order-1 area can be located far from the farthest one-hop neighbors. Thus
but order-2. So area_gw is aggregated into order-2 area chances are high to find an area closer to the destination.
that is area_2. Then the added entry is Secondly, the one-hop broadcast and non geographical nature
{Order-2 aggr., area_2, GW, C0 }. of the the updating mechanism allows the routing information
to circumvent local maxima. Thus some forwarding candidates
E. Forwarding areas are made reachable even through gateways that are at
Forwarding decisions rely on information contained in the greater distance from the destination than the source.
routing table. Example. On Fig.5, the node N has to forward a packet
When a node receives a packet, first it checks if it is the whose final destination is D. In its routing table it has the
destination. If not, it has to choose a node to which the packet following entries among others:
will be forwarded. This process is repeated by every node until • {Order-1 aggr., area_1, A, C}: A is the gate-

the packet reaches the final destination. way to reach the area area_1;
We recall the assumption that the sender node is aware of the • {Order-2, area_2, B, C}: B is the gateway to

geographical location of the destination either inferred from its reach the area area_2.
ID or from a location service [9]. Also, each aggregate area Because area_2 is closer to D than area_1, the packet will
has a defined geographical location corresponding to that of be forwarded to B even if B is farther from the destination than
its reference point. The reference point of an aggregate area N.
can be defined as its geometric centroid. F. Discussion
We propose a greedy-like forwarding process. In a pure
greedy routing [4], the next hop for a forwarded packet is one In this section we evaluate qualitatively our solution by
of the one-hop neighbors nodes that is closer to the destination. discussing its performance according to some metrics.
In our protocol, the next hop is chosen as the gateway to the 1) Table size: The routing table size (the number of entries
node or the area (regarding its reference point) that is closest inside tables) has a direct impact on the amount of memory
to the destination. We use euclidean distances for proximity consumed by every node in order to run the protocol properly.
comparisons. In classical table-based routing protocols, for a connected
Algorithm. Let N a node receiving a packet whose final network, the table size is linear with the number of nodes in
destination is D. the network as ultimately each node will trigger an entry. Our
proposition reduces the size of tables thanks to the concept
• Select all the gateways offering a node or an area closer
of areas which aggregates node inside them. Furthermore,
to D: smaller tables leads to smaller route updates meaning reduced
– If no gateway is found, forwarding fails; overhead.
– Else, among the selected gateways, choose the one 2) Convergence time: Given a static network where all
offering the node or the area closest to D. nodes start at the same time, the convergence time is the
Pure greedy routing fails when no neighbor is at a smaller necessary amount of time for each node’s routing table to
distance to the destination (local maximum). In our solution, stabilize. Compared to classical table-driven routing [10], our
forwarding fails when all the areas reachable through the solution needs shorter time to converge. In fact, a routing
 The local area of a node is implemented as the geographical
region formed by the order-1 aggregate area containing the
2 2
node and all its adjacent order-1 areas (Fig.6a).
3 Order-n aggregate areas are implemented as order-n areas
1 1
2 adjacent to the order-n area containing the node. Fig.6b shows
N N 1
examples of aggregate areas with their respective orders.
1 1
2 2

3 3
C. Simulation parameters
We conducted our simulation using ns-3 (ns-3.24 release).
Each node implements the 802.11b model with 2 Mbps rate
(a) Local area (b) Aggregate areas and 250 m range. The nodes are randomly deployed in the
Fig. 6. Local area and aggregate areas network zone and remain static. They are powered on at
the same time. For routing table size and convergence time
measurements, we simulated 20 to 300 nodes with constant
update leads to table modification in a limited range of nodes. density of 1 node per 9000 m2 . For PDR scenario, we
As seen previously, when a node receives a route update, it simulated 200 nodes with density varying from 0.25 node per
looks for an aggregate area for the destination. The further 9000 m2 to 1.25 nodes per 9000 m2 . The nodes are deployed
the destination is, the higher the probability that the aggregate in variable size square regions to reflect the density. 10 nodes
area is a big one. In this case, chances are high that a route are selected randomly to generate 2 kbps CBR flows with 64
with the same aggregate area has already been received, so no bytes packets sent to other 10 random nodes. The flows are
update occurs. started after the convergence of the routing tables and last
3) Packet delivery ratio (PDR): In the case of static net- for 10 seconds. There is no loss due to interference between
work, our proposition is expected to be a tradeoff between a flows as they are not concurrent. We chose an interval of 15
full table-driven routing protocol and a geographical one. More s between routing periodic updates. A simulation ends 10 s
precisely we are improving the table-driven aspect by reducing after the last sent packet.
the table size and bandwidth consumption (as seen before)
with the cost of lower PDR because of routing information D. Results
aggregation. But the use of routing tables can circumvent some
of the local maxima making our solution more efficient than 1) Routing table size: The Fig.7a shows the average num-
pure geographical routing protocol. ber of entries in the routing tables with different network sizes.
4) Mobility: The study of mobility needs extensive simula- As we can see, the average number of routing table entries
tion and is out of the scope of this paper. However we wanted for increasing nodes number has a logarithmic shape. This
to highlight a potential advantage of our solution. Thanks to shows the scalability of our proposition as the average routing
the aggregate areas and routing tables having areas as entries, table size doesn’t explode with large number of nodes in the
we tolerate some local mobility of distant nodes. In fact, for network.
a given node, mobility of other nodes that are confined inside 2) Convergence time: The convergence time is the nec-
an aggregated area don’t trigger any update. In other words, essary amount of time for all the nodes tables to stabilize
they mostly generate updates at a local level. after start-up for a static network. Before convergence, the
tables have incomplete routing information resulting in poor
III. S IMULATION forwarding decisions. Thus the convergence time has an im-
A. Objectives pact on how fast a network can be deployed. It is interesting
The main goals are to show that our proposition is more to see how this metric evolves with the size of the network as
efficient compared to pure geographical routing while having short deployment time even with high number of nodes can
good scaling capabilities in terms of routing table size and be considered as a criteria of scalability. The Fig.7b shows
convergence time. We measured table size, convergence time that the convergence time grows slowly with the number of
and packet delivery ratio (PDR) in regard to nodes number deployed nodes.
and nodes density. 3) Packet delivery ratio (PDR): The PDR is the ratio
between the number of packets successfully received by the
B. Local area and aggregate areas implementation destination and the total number of packets sent by the
In order to partition the network zone, we chose the hi- source. The Fig.7c shows the average PDR with different
erarchical area partitioning proposed in [11]. Their algorithm node densities. As we can see, with a large range of node
consists in a recursive subdivision of the whole network zone densities, our proposition brings substantial improvement to
into square areas: 4 order-n areas form one order-n + 1 area. the PDR compared to pure greedy routing. It demonstrates
The size of the primary areas (order-1) is chosen such as two its capacity for circumventing the local maxima, reducing
nodes inside the same order-1 area can reach each other. forwarding failures.
 60 50 100
Average # of entries

Convergence (s)
50 40 90
40 80

PDR (%)
30
30 70
20
20 60 Our proposition
10 10 50 Greedy
0 0 40
20 50 100 150 200 250 300 20 50 100 150 200 250 300 0.25 0.5 0.75 1 1.25

# of nodes # of nodes Density (x*1/9000)

(a) Table size (b) Convergence time (c) PDR

Fig. 7. Simulation results

IV. R ELATED WORK R EFERENCES

In [10] Mohseni et al. propose a comparative study of [1] F. Dressler, “Self-Organization in Ad Hoc Networks: Overview and
topology-based routing protocols. For the position-based ap- Classification,” University of Erlangen, Dept. of Computer Science 7,
Technical Report 02/06, March 2006.
proaches, a comprehensive survey can be found in [4]. To [2] S. A. R. Zekavat, S. Kansal, and A. H. Levesque, “Wireless positioning
avoid local maximum handling, some geographical routing systems: Operation, application, and comparison,” in Handbook of
protocols construct landmarks [12] or spanning tree-like struc- Position Location. John Wiley & Sons, Inc., 2011, pp. 3–23.
[3] D. Chen and P. K. Varshney, “A survey of void handling techniques
ture [13]. The idea of using areas for routing purpose has been for geographic routing in wireless networks,” IEEE Communications
already proposed. ZRP [14] defines two areas for applying Surveys Tutorials, vol. 9, no. 1, pp. 50–67, First 2007.
respectively reactive and proactive topology-based routing [4] F. Cadger, K. Curran, J. Santos, and S. Moffett, “A survey of geographi-
cal routing in wireless ad-hoc networks,” IEEE Communications Surveys
protocol. Another protocol using areas not in a geographical Tutorials, vol. 15, no. 2, pp. 621–653, Second 2013.
way is DART [15] which uses its own adressing scheme [5] M. Mauve, J. Widmer, and H. Hartenstein, “A survey on position-based
to define logical hierarchical areas. In a more geographical routing in mobile ad hoc networks,” IEEE Network, vol. 15, no. 6, pp.
30–39, Nov 2001.
fashion, [16] proposes region summaries of nodes with routing [6] L. Blažević, S. Giordano, and J.-Y. Le Boudec, “Self organized termin-
on them. Regarding variable routes details, FSR [17] takes into ode routing,” Cluster Computing, vol. 5, no. 2, pp. 205–218, 2002.
account the proximity of nodes to each other to adapt routing [7] D. D. Couto and R. Morris, “Location proxies and intermediate node
forwarding for practical geographic forwarding,” MIT Laboratory for
updates precision. Computer Science, Tech. Rep., June 2001.
V. C ONCLUSION [8] C. E. Perkins and P. Bhagwat, “Highly dynamic destination-sequenced
distance-vector routing (DSDV) for mobile computers,” in SIGCOMM
We propose in this paper, a geographical routing protocol ’94 Conference on Communications Architectures, Protocols and Appli-
provided with topological information. Every node maintains cations, 1994, pp. 234–244.
[9] S. Das, H. Pucha, and Y. Hu, “Performance comparison of scalable lo-
a routing table containing hosts entries for closer nodes and cation services for geographic ad hoc routing,” in INFOCOM 2005 24th
areas entries aggregating further nodes. To make it possible Annual Joint Conference of the IEEE Computer and Communications
for the node to aggregate and exchange nodes information, the Societies, vol. 2, March 2005, pp. 1228–1239.
[10] S. Mohseni, R. Hassan, A. Patel, and R. Razali, “Comparative review
deployment area is splitted into partitions. Routing decisions study of reactive and proactive routing protocols in MANETs,” in 4th
are taken solely on routing table contents: a packet is sent to IEEE International Conference on Digital Ecosystems and Technologies,
the gateway of entry nearest to the destination. 2010, pp. 304–309.
[11] M. N. Ranaivo Rakotondravelona, F. Harivelo, and P. Anelli, “Hierar-
Evaluation compares the proposition performances with chical area-based address autoconfiguration protocol for self-organized
those of a pure geographical routing. It shows higher Packet networks,” in Ad Hoc Networks - 7th International Conference, AdHoc-
delivery ratio validating that the proposed protocol makes a Nets 2015, Proceedings, 2015, pp. 173–184.
[12] J. Na and C. kwon Kim, “GLR: A novel geographic routing scheme for
better use of link information by reducing forwarding fail- large wireless ad hoc networks,” Computer Networks, vol. 50, no. 17,
ures. The size of the routing table and the convergence time pp. 3434 – 3448, 2006.
increase slowly in logarithmic shape with the number of nodes, [13] B. Leong, B. Liskov, and R. Morris, “Geographic routing without
planarization.” in NSDI 2006 3rd Symposium on Networked Systems
demonstrating a good scale factor stability. Design and Implementation, vol. 6, 2006, p. 25.
The way our proposal introduces topological information in [14] B. Patel and S. Srivastava, “Performance analysis of zone routing
location-based routing, can be seen as widening the possibility protocols in mobile ad hoc networks,” in NCC 2010 National Conference
on Communications, Jan 2010, pp. 1–5.
for a node to choose the next hop. While on classical geo- [15] J. Eriksson, M. Faloutsos, and S. Krishnamurthy, “DART: Dynamic
graphical routing, the next hop is a node, in our proposition, address routing for scalable ad hoc and mesh networks,” IEEE/ACM
the next hop can be considered as an area: geographical routing Transactions on Networking, vol. 15, no. 1, pp. 119–132, Feb. 2007.
[16] P.-H. Hsiao, “Geographical region summary service for geographical
is then extended to areas. Future work will study quantitatively routing,” SIGMOBILE Mobile Computing and Communications Review,
the impact of mobility on protocol performance. vol. 5, no. 4, pp. 25–39, Oct. 2001.
[17] A. Iwata, C.-C. Chiang, G. Pei, M. Gerla, and T.-W. Chen, “Scalable
ACKNOWLEDGMENTS routing strategies for ad hoc wireless networks,” vol. 17, no. 8, pp.
This research was supported by Région Réunion and the 1369–1379, 1999.
European Union (ERDF).