Вы находитесь на странице: 1из 7

Optimal Path Finding Using Sequential Intermediate Destinations and

Randomized Routing Algorithms

Harvey Ian Arbas1,2 a


, John Justine Villar1,2 b
1 Scientific Computing Laboratory, Department of Computer Science, University of the Philippines - Diliman, Philippines
2 Intelligent Transportation Systems Laboratory, National Center for Transportation Studies, Philippines
hsarbas@up.edu.ph, jsvillar1@up.edu.ph

Keywords: Randomized routing algorithms, Sequential intermediate destinations, Optimal path finding

Abstract: Routing algorithms are protocols which are most commonly used in efficiently directing internetwork traffic.
There are a number of deterministic and randomized routing algorithms proposed in existing literature, partic-
ularly in mobile ad hoc networks. In this paper, we take these routing algorithms and test their effectiveness
when implemented in road transport networks. Our simulation results show that internet routing protocols
are effective in finding an optimal path in a road network given an origin-destination pair, and a set of land-
mark nodes referred to as intermediate nodes. Results also show that implementing sector-based algorithms
significantly improves the path finding ability of these algorithms.

1 Introduction eler’s perspective the set of all possible paths from


origin to destination are filtered by the traveler’s
Routing algorithms are step-by-step operations own perception, cognition, and idiosyncrasies (Bovy,
used in directing internet traffic effectively and ef- 2009). These filters are simplified into constraints
ficiently in a network or across multiple networks. which attempt to approximate the traveler’s tendency
When a data packet leaves its source node, the rout- to choose reasonable routes (Kaplan and Prato, 2012).
ing algorithm determines which path to take to reach Bovy and Stern (1990) also defined a spatial restric-
its destination node. In wireless networks, a mobile tion and directionality of travel considering the desti-
ad hoc network (MANET) is a collection of inter- nation. This space-time constraint commonly comes
communicating mobile devices that do not need any in the form of an ellipse with the origin and destina-
fixed infrastructure and can be represented by a graph, tion nodes as its foci (Bovy and Stern, 1990).
where two nodes are connected if and only if their An alternative perspective on the route choice pro-
Euclidean distance is at most a given unit disk radius cess for travelers incorporated the travelers’ behavior
(Barrire, Fraigniaud, Narayanan, and Opatrny, 2003). to establish a mental map through perceiving, storing,
Several routing protocols have been devel- and remembering landmarks, referred to as anchor
oped and implemented for MANET; Deterministic points. These anchor points are treated as intermedi-
progress-based routing algorithms such as greedy ate destinations by the traveler and are successively
routing and compass/directional routing are known to reached until finally reaching the final destination,
fail to deliver the packet in certain conditions (Karp thus a sequential approach to route choice (Gentile
and Kung, 2000). To improve the delivery rates of and Papola, 2006). A method of generating these in-
these algorithms, randomized algorithms were also termediate destinations was proposed using road hier-
proposed (Fevens, Haque, and Narayanan, 2005). archies as basis for computing a node’s nodal affinity
In traffic and transportation, efforts have been score (Carigma, Eden, Villar, and Valdez, 2019). This
made to model route choice behavior. Route choice node scoring method determines whether a node is a
behavior described in (Bovy, 2009) and (Bovy and strong candidate to be an intermediate destination for
Stern, 1990) model route choice from the perspec- a given road network.
tives of both traveler and modeler. From a trav- This paper studies effectiveness of implementing
routing protocols, particularly the randomized rout-
a https://orcid.org/0000-0003-0586-7773 ing algorithms, in conjunction with the elliptical and
b https://orcid.org/0000-0003-1617-4518 sequential intermediate destinations models to deter-
mine an optimal path in a road transport network Minimality Profitable/minimal routing algorithms
given an origin node and a destination node. This always give channel that bring the packet closer to its
will be implemented in a hypothetical grid network destination while misrouting/nonminimal routing al-
applied with the node scoring method. gorithms also allow the packet to move away from its
The rest of the paper is structured as follows. Sec- destination.
tion 2 presents relevant literature that were crucial
in the implementation of this study. In Section 3, a Number of paths Routing algorithms may also be
detailed discussion of the simulation environment in classified according to the number of alternative paths
which the models were implemented is presented fol- as either completely or partially adaptive.
lowed by a summary and analysis of the results in
Section 4. Finally, conclusions and further research 2.2 Progress-based algorithms
directions are given in Section 5.
Progressive routing algorithms, also called progress-
based algorithms, (Giordano and Stojmenovic, 2004),
2 Related Literature forward the packet to a neighbor node, which is cho-
sen according to a specified heuristic. One type of
progress-based algorithm is the greedy routing where
2.1 Routing algorithms a node forwards the packet to the neighbor closest to
the destination (Stojmenovic and Xu Lin, 2001; Finn,
In interconnection networks, messages are sent from 1987). Compass or directional routing on the other
one node to another through a path consisting of a se- hand moves the packet to a neighboring node such
ries of network links. These paths are established by that the angle formed between the current node, next
routing algorithms. Routing algorithms can be clas- node, and destination node is minimized (Kranakis,
sified according to several criteria (Duato, Yalaman- Singh, and Urrutia, 1999). Clearly, the next node se-
chili, and Lionel, 2002): lected by these two algorithms may not always be the
same (see Figure 1 for an example).

Number of destinations Messages may have a sin-


gle destination (unicast routing) or multiple destina-
tions (multicast routing).

Routing decisions Routing decisions can be estab-


lished by a centralized controller (centralized rout-
ing), at the source node prior to packet injection
(source routing), or in a distributed manner (dis-
tributed routing). Hybrid schemes, called multiphase
routing, are also possible.

Implementation Routing algorithms can be imple-


mented using a routing table (table lookup) or execut-
ing according to a finite-state machine.
Figure 1: Greedy chooses E while Compass chooses F as
the next node. Lifted from (Fevens et al., 2005).
Adaptivity When choosing a path between a given
source/destination pair, routing algorithms may al-
ways output the same path (deterministic) or use in- 2.3 Randomized routing algorithms
formation about the current network traffic to avoid
congested regions (adaptive). Progress-based routing algorithms, both deterministic
and randomized were applied to mobile ad hoc net-
works. A routing algorithm is randomized if the next
Progressiveness Progressive routing algorithms al- node is chosen randomly out of the neighbors of the
ways move the header forward as opposed to back- current node (Bose and Morin, 1999). The study then
tracking algorithms which allow the header to back- compared the algorithms’ performance in terms of de-
track, releasing previously reserved channels. livery rate, the percentage of times that the algorithms
succeeds in finding a path, and stretch factor, the av- next node is chosen uniformly at random out of these
erage ratio of the length of the path found by the algo- two nodes.
rithm to the length of the shortest path in the network.
Two of the algorithms presented were deterministic. Best2Compass Let n1 and n2 be the neighbors of
Of the remaining seven randomized algorithms, three c such that ∠dcn1 (or ∠n1 cd) and ∠n2 cd (or ∠dcn2 )
were sector-based. are the two smallest such angles among all neighbors
Given a node u, the disk centered at node u with of c. The next node is chosen uniformly at random
radius l is denoted by disk(u, l). Given an angle θ out of n1 and n2 .
such that 0 ≤ θ ≤ 2π, Sector(c, d, θ) is defined as the FarInSector, another sector-based algo-
sector given by the angle θ in disk(c, r) that is bisected rithm which uses additional concepts such as
by the line segment cd (Fevens et al., 2005). Figure 2 Periphery(c, d, θ, α),Core(c, d, θ, α), andWing(c, d, θ, α),
shows an illustration of the definition of Sector. was not implemented in this study.

2.4 Detour factor

A transportation network can be reduced to a reason-


able subnetwork or a potential path area using an
elliptical construct as presented in (Ma, Leite Mari-
ante, and Van Acker, 2017). The study showed that
the boundary of the ellipse maintain a constant de-
tour factor, which is a measure of deviation from the
origin-destination axis. The detour factor, defined as
a+b
c , sets a threshold to limit the path taken by the
traveler. This is shown in Figure 3.
Figure 2: Illustration of the definition of Sector. Lifted from
(Fevens et al., 2005).

GreedyInSector If Sector(c, d, θ) is not empty, the


next node x is chosen to be the neighbor of c having
minimum distance from d, in Sector(c, d, θ).

RandomInSector The next node x is chosen uni-


formlu at random from the first non-empty set in the
following sequence: Sector(c, d, θ1 ), Sector(c, d, θ2 ),
disk(c, r), where 0 ≤ θ1 ≤ θ2 ≤ 2π.
Figure 3: Ellipse with origin o and destination d as the foci.
RCompass Let ∠xuy denote the angle formed by x, Lifted from (Ma et al., 2017).
u, and y measured counterclockwise. Let n1 be the
neighbor of c above the line cd such that ∠dcn1 = θ1
is the smallest among all such neighbors. Similarly, 2.5 Intermediate destinations
n2 is the neighbor of c below cd such that ∠n2 cd =
θ2 is the smallest among all such neighbors. The next In a road network, a node is deemed more desir-
node x is chosen uniformly at random from n1 and n2 . able if roads of higher hierarchy are directly acces-
sible since it provides the traveler more alternatives
to travel along faster corridors (Bovy, 2009). To im-
WeightedRCompass Let n1 and n2 be the neigh-
plement this, the concept of nodal affinity score was
bors of c as defined in RCompass. The next node x is
introduced (Carigma et al., 2019). From the set of
chosen from n1 and n2 with probability θ2 /(θ1 + θ2 )
all nodes excluding the origin node and destination
and θ1 /(θ1 + θ2 ), respectively.
node, a subset of candidate intermediate nodes are se-
lected and given a nodal affinity score . This is the
Best2Greedy Let n1 and n2 be the closest and sec- score attributed to a node based on the hierarchy of
ond closest neighbors of c to the destination d. The the links connected to it. A test network modeled as
Table 1: Nodal affinity score and speed limit per road hierarchy (Carigma et al., 2019).
Road Hierarchy Nodal Affinity Score Road Speed Limit (kph)
Service Road 1.00 40
Residential Road 2.00 50
Tertiary Road 2.50 60
Secondary Road 2.83 70
Primary Road 3.08 80
Trunk 3.28 90
Motorway 3.45 100

a 2D graph with the edges representing roads of vary- (node labels start from the bottom-left at (0, 0)). Ori-
ing hierarchies was presented. A list of the road hier- gin and destination nodes were fixed at (0, 0) and
archies with their corresponding scores along with an (n − 1, n − 1) respectively for all runs. Finally, all
the speed limit associated with the road hierarchy was roads in the network were assigned random hierar-
also presented in the study (see Table 1). chies.
After representing the given road transport net-
work as a nxn graph, a subraph containing the po-
tential path area described in Section 2.4 is gener-
3 Methodology ated. As in (Carigma et al., 2019), the detour fac-
tor was assumed to be an inverse function of the dis-
3.1 Simulation environment tance between the given origin and destination nodes,
instead of the corresponding travel time. Using the
method described in Section 2.5, nodal affinity scores
for each node were calculated. The set of candidate
intermeiate nodes was then generated by selecting the
200 top-scoring nodes. Figure 4 illustrates the graph
representing the road network, the ellipse determined
by the detour factor, and the set of candidate interme-
diate nodes for the given origin and destination nodes.

Figure 4: Road network represented as a graph. Origin and


destination nodes are colored blue, nodes inside the detour
factor are colored yellow, and candidate intermediate nodes
are colored green.

In the simulation experiments, a road network is


represented by a nxn grid where n ∈ {25, 50, 75, 100}.
Each edge in the graph represents a road, with edge Figure 5: Optimal path generated by the algorithm is col-
weight equal to the length of the road. For simplic- ored red, actual shortest path is colored orange.
ity, each road was assigned a length randomly cho-
sen from [1m, 10m]. Moreover, each node was as- With the exception of FarInSector, all algorithms
signed a label corresponding to its row and column discussed in Section 2.3 were implemented. However,
Table 2: Average stretch factor and standard deviation, σ.
n = 25 n = 50 n = 75 n = 100
Algorithms ave. σ ave. σ ave. σ ave. σ
Greedy 1.09 0.06 1.13 0.09 2.83 3.27 7.67 3.59
GreedyInSector 1.09 0.06 1.13 0.09 1.19 0.03 1.25 0.08
RandomInSector 1.20 0.11 1.23 0.12 1.31 0.08 1.36 0.05
Compass 1.10 0.05 1.14 0.07 3.05 3.76 7.82 3.75
RCompass 1.64 0.84 3.49 2.23 6.38 3.09 8.78 2.71
WeightedRCompass 1.10 0.05 1.28 0.30 3.24 3.96 7.87 3.69
Best2Compass 1.11 0.06 1.14 0.09 2.91 3.44 8.41 3.92
Best2Greedy 1.12 0.07 1.16 0.08 3.05 3.68 7.18 3.87

some modifications to the algorithms were made. As 4.2 Average stretch factor
in (Fevens et al., 2005), transmission radius was ini-
tially set to 15m. In the event that an algorithm is un- The average stretch factor of each algorithm is pre-
able to select the next intermediate node given its cur- sented in Table 2. From the results we can clearly see
rent transmission radius, the radius is incremented by that the stretch factor for all the algorithms increases
one meter and the process of selecting the next node as n increases. Moreover, the deterministic routing
is restarted. This step is repeated until the algorithm algorithms (Greedy, GreedyInSector, and Compass)
successfully finds the next viable intermediate node. performed better than the randomized algorithms in
This ensures that a path is found from the given origin general.
and destination pair. A shortest path is then generated It is important to note that the sector-based routing
from the current node to the determined intermediate algorithms (GreedyInSector, RadnomInSector) per-
node. The whole process is then repeated until the formed significantly better than all the non sector-
destination node is reached. The optimal path is con- based algorithms, both deterministic and random-
catenated shortest paths generated by the algorithm. ized, especially on networks having higher number of
Figure 5 illustrates the path given by the algorithm nodes. This is because the Sector restricts the extent
and the actual shortest path in the network. to which the path strays away from the directoin of
For each of the network sizes given, each algo- the destination. Furthermore, the deterministic sector-
rithm was ran 10 times. Average delivery rates and based GreedyInSector algorithm performed slightly
average stretch factors were then computed. A tabu- better than the randomized sector-based RandomIn-
lation and interpretation of the results is presented in Sector algorithm. A sample run of the GreedyInSec-
the next section. tor algorithm in different test networks is shown in
Figure 6.

4 Results and Discussion


5 Conclusion and Future Research
We are interested in the performance of the algo- Directions
rithms presented in the literature. In particular, we
want to see how each algorithm behaves as the num- In this paper, we studied the effectiveness of im-
ber of nodes in the network increases. plementing routing protocols normally used in rout-
ing internet traffic into a road transport network in
4.1 Average delivery rate determining an optimal path given an origin node, a
destination node, and a set of intermediate nodes. We
In (Fevens et al., 2005), a run is considered success- compared the performance of several deterministic
ful if a path from the origin to the destination is dis- and randomized routing algorithms proposed in ex-
covered. However, since the modifications mentioned isting literature. In particular, we compared the path
above guarantees that a path is found, we are sure determined by the algorithms with the actual shortest
to get 100% delivery rates for all implemented algo- path in the network by computing for its stretch fac-
rithms. This result is justified on the basis that the tor. Results show that the value of the stretch factor
traveler should be able to reach their destination re- increases as the number of nodes in the network in-
gardless of how far the next intermediate destination creases. It was also found out that the sector-based al-
is from the current node. gorithms performed significantly better than their non
(a) n = 25 (b) n = 50

(c) n = 75 (d) n = 100


Figure 6: Sample run using GreedyInSector in different nxn grid networks.

sector-based counterparts. in Triangulations. In Algorithms and Computa-


Finally, the results of this paper show that rout- tion. Springer Berlin Heidelberg, Berlin, Hei-
ing protocols can in fact be used in road transport delberg, 113–122.
networks. We are now interested in the implementa- P.H. Bovy and E. Stern. 1990. Route Choice:
tion of more routing protocols which can find optimal Wayfinding in Transport Networks. Springer
paths with lower stretch factors. Netherlands.
Piet H. L. Bovy. 2009. On Modelling Route
Choice Sets in Transportation Networks:
References A Synthesis. Transport Reviews 29, 1
(2009), 43–68. https://doi.org/10.1080/
Lali Barrire, Pierre Fraigniaud, Lata Narayanan, and 01441640802078673
Jaroslav Opatrny. 2003. Robust position based Anjilo Carigma, Mel Francis Eden, John Justine Vil-
routing in wireless ad hoc networks with unsta- lar, and Adrian Roy Valdez. 2019. Sequential
ble transmission ranges. Wireless Communica- Intermediate Destination Approach to Route
tions and Mobile Computing 3 (03 2003), 141– Choice Set Generation. In Proceedings of the
153. https://doi.org/10.1002/wcm.108 Eastern Asia Society for Transportation Stud-
Prosenjit Bose and Pat Morin. 1999. Online Routing ies, Vol.12,2019. 18.
Jose Duato, Sudhakar Yalamanchili, and Ni Lionel.
2002. Interconnection Networks: An Engineer-
ing Approach. Morgan Kaufmann Publishers
Inc., San Francisco, CA, USA.
T. Fevens, I. T. Haque, and L. Narayanan. 2005.
Randomized Routing Algorithms in Mobile Ad
Hoc Networks. In Mobile and Wireless Com-
munication Networks, Elizabeth M. Belding-
Royer, Khaldoun Al Agha, and Guy Pujolle
(Eds.). Springer US, Boston, MA, 347–357.
G.G. Finn. 1987. Routing and addressing prob-
lems in large metropolitan-scale internetworks.
(1987).
Guido Gentile and Andrea Papola. 2006. AN ALTER-
NATIVE APPROACH TO ROUTE CHOICE
SIMULATION: THE SEQUENTIAL MOD-
ELS. (01 2006).
Silvia Giordano and Ivan Stojmenovic. 2004. Po-
sition Based Routing Algorithms for Ad Hoc
Networks: A Taxonomy. Springer US, Boston,
MA, 103–136. https://doi.org/10.1007/
978-1-4613-0223-0 4
Sigal Kaplan and Carlo Prato. 2012. Closing the gap
between behavior and models in route choice:
The role of spatiotemporal constraints and la-
tent traits in choice set formation. Transporta-
tion Research Part F: Traffic Psychology and
Behaviour 15 (01 2012), 924. https://doi
.org/10.1016/j.trf.2011.11.001
Brad Karp and H. T. Kung. 2000. GPSR: Greedy
Perimeter Stateless Routing for Wireless Net-
works. In Proceedings of the 6th Annual Inter-
national Conference on Mobile Computing and
Networking (MobiCom ’00). ACM, New York,
NY, USA, 243–254. https://doi.org/10
.1145/345910.345953
Evangelos Kranakis, Harvinder Singh, and Jorge Ur-
rutia. 1999. Compass Routing on Geometric
Networks. In IN PROC. 11 TH CANADIAN
CONFERENCE ON COMPUTATIONAL GE-
OMETRY. 51–54.
Tai-yu Ma, Gabriel Leite Mariante, and Veronique
Van Acker. 2017. Location choice modeling
based on mixed logit model and sampling of al-
ternatives.
I. Stojmenovic and Xu Lin. 2001. Loop-free hy-
brid single-path/flooding routing algorithms
with guaranteed delivery for wireless net-
works. IEEE Transactions on Parallel and Dis-
tributed Systems 12, 10 (Oct 2001), 1023–1032.
https://doi.org/10.1109/71.963415

Вам также может понравиться