Hierarchical Adaptive Location Service Protocol for ... - Semantic Scholar

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

Hierarchical Adaptive Location Service Protocol for Mobile Ad Hoc Network Sabbir Ahmed, Gour C. Karmakar and Joarder Kamruzzaman Monash University, Australia {sabbir, gour, joarder}@infotech.monash.edu.au Abstract— Position based routing protocols have lower routing overhead due to exploiting position information of mobile nodes for forwarding data. The performance of location based protocols depends on the precise knowledge of the destination’s location. Therefore a location service is a prerequisite, from which a transmitter can find the approximate location of the receiver node. Several location service schemes have been proposed in literature, among them hierarchical services became attractive due to their scalability. These schemes adopt hash function based location server (home) assignment which requires nodes to be distributed throughout the concerned area uniformly. Node mobility in real world may cause non-uniform node distribution under which condition performance of the existing location schemes degrades considerably. This demands an improved location service scheme which can adapt itself with all contextual situations. In this paper we propose a novel location service scheme which performs better than existing location services in both uniform and non-uniform node distributions while maintaining scalability in location update and query. Keywords-location service; geographic routing;

I.

INTRODUCTION

Over the last decade, huge research efforts have been dedicated to a challenging research issue - to design scalable routing protocol. Despite there are many types of routing protocols exist in literature the most promising group of routing protocols is position based routing protocol, which includes Distance Routing Effect Algorithm for Mobility (DREAM) [1], Location-Aided Routing (LAR) [2] and Greedy Perimeter Stateless Routing (GPSR) [3]. These protocols use geographic location of destination node to forward data. Routing protocols other than location-based maintain routes either proactive, on demand or hybrid manner. However, if the network size increases these approaches fail to scale well due to their nature of route discovery. In addition, increased mobility results higher resource consuming route computation. With the availability of GPS and other relevant techniques, a mobile device can easily find its own location. Nodes can achieve their neighbors’ location through periodic HELLO message. This enables a source to choose a neighbor which is nearest to the destination among its neighbor list. This greedy approach fails if there is no such neighbor in the direction of destination. In this situation, GPSR recovers by forwarding in perimeter mode. Since the approximate location of destination is needed in geographic routing protocols, the presence of location service is mandatory and the location service must not be a burden to the routing protocol itself. By broadcasting query a node may achieve another nodes location but requirement of broadcasting makes this method not suitable. To overcome the overhead and to increase scalability in location update and query cost, for

instance a node A selects a subset of nodes as its location servers either forming quorums or using uniform hash function to store its location and in a similar way any other node willing to know A’s location can send query to those servers. Different types of server selection and maintenance schemes are available which are described in details in Section II. Although many location service schemes have been proposed for MANET, hash function based hierarchical organized location services show better scalability in terms of location update and location query cost. One of the inherent characteristics of hash function based server assignment is to divide the whole area into square grids and each square is assigned uniformly as home region for nodes based on their IDs. For this reason every node in a specific region becomes responsible for acting as location server for some nodes. Since every square is a home of subset of nodes, due to the node mobility it may happen that there is no node in that particular cell. This empty home region causes both update and query failure for those nodes it has been assigned as home. Several schemes try to recover by storing the update information and forwarding the query around that cell. Though this recovery process may serve the purpose for short time empty home handling but if the cell becomes empty for a long time or a larger area containing multiple cells become empty, the recovery process may not be suitable and increases the overhead for the nodes creating perimeter. For this reason, when the nodes become non-uniformly distributed due to the nature of different mobility models, the performance is expected to degrade considerably. So, handling non-uniform node distribution should be kept in mind while designing location service since in real world non-uniform node distribution is common due to the presence of hot spot or region of interest. In this context, we propose a novel location service protocol that can handle both uniform and non-uniform node distribution. For scalability reason, we employ hierarchy in grid structure, but do not use hash based approach. Moreover a home region shifting mechanism is proposed which efficiently manages situations arising from empty home regions. It can also adaptively create and destroy home region when required. Furthermore, we implemented our proposed location service in NS2 and simulation results show superior outcome of our proposed model over an efficient and scalable hierarchical location service protocol. II.

RELATED WORKS

In literature various location service protocols are found. These protocols can be classified in four categories: Flat, Two level, Multi-level and Quorum based structure. One of the very first location services found in the literature is Scalable Location Update-Based Routing Protocol (SLURP)[4]. In this

978-1-4244-2948-6/09/$25.00 ©2009 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

method, the network area is divided into a flat grid of squares. For each arbitrary node A, a hash function is applied on its ID that maps a specific geographic region, called as home region for A. Nodes residing in that region becomes location servers for A. In a similar way, when another node wants to know the location of A, it uses the same hash function and sends a query. This flat structure suffers from a well-known problem called distance-effect- even two nodes A and B are geographically close to each other but query may need to travel long distance. In order to overcome the distance effect two-level structure was introduced in Scalable Ad-hoc Location Management (SLALoM) [5], Distributed Location Management (DLM) [6] and Adaptive Demand-driven Location Service (ADLS) [7]. SLALoM combines the concept of two-level hierarchical grid and SLURP by introducing near and far home region. ADLS follows similar strategy like SLALoM but level-2 home region is generated on demand basis. Two level structure reduces query cost but increases update cost significantly. The first location service protocol belonging to multi-level hierarchical structure is Grid Location Service (GLS) [8]. GLS divides the area into a hierarchy of squares forming a quad-tree. Each node selects one node in each region of the quad-tree as a location server. Therefore the density of location servers for a node is high in areas close to the node. Hierarchical Location Service (HLS) [9] and HIerarchical Geographical Hashing with multi-GRained Address DEelegation (HIGH-GRADE) [10] follow similar strategy for selecting home region and query forwarding. At the top level, the entire area is called a level-H square, where H denotes the number of levels in the hierarchy and is divided into four quadrants recursively, until 4H level 0 squares. For an arbitrary node A’s level-i servers store information on which level (i-1) square A resides in. Only level-0 servers store A’s exact location. When a node crosses a level i region it updates only its level (i+1) home region thus reducing the update cost. When a node B is willing to find A’s location it uses hash function on A’s ID and if A’s location is not found in there, the query is forwarded to the next (i+1)th level home region recursively until the hth (h ≤ H) level is reached which is the smallest common region where both A and B resides in. One different strategy of developing location service found in literature is quorum based service. In quorum based system, a node selects a set of nodes that produce a quorum for holding that nodes location information. Scalable quorum-based location service [11], SEEKER [12] and Locality aware Location Service (LLS) [13] fall in this classification. In quorum based system each node updates its location in a specific direction either north-south, or east-west, or spiral like structure and in a similar manner location query is initiated. Nodes residing at where update and query path intersects are expected to know the location of queried node. But as mentioned in [11] quorum-based scheme may also suffer from distance effect and attractive when nodes move more or less the same direction keeping the relative position same to avoid frequent location update. There exists a design tradeoff [10] for reducing update and query cost but among the above mentioned location services multi-level structure location services show a right balance in update and query cost while maintaining better scalability over others. Thus we become motivated to propose a new location service belonging to this category.

III.

HIERARCHICAL ADAPTIVE LOCATION SERVICE (HALS)

In this section we present the theoretical underpinnings of our proposed location service protocol, namely Hierarchical Adaptive Location Service (HALS). A. Generation of grids Like other contemporary and efficient hierarchical location services (HIGH_GRADE, HLS) HALS also divides the whole region into grids. The smallest square regions are termed as level-0 cells (L0). After generating the L0 cells multiple cells are grouped hierarchically into higher level cells. Since empty home region reduces update and query success rate in this model we intend to shift location server (home region) so that home region does not become empty. Therefore, the L1 cell is composed of a higher number of L0 cells, so that shifting mechanism does not cause overhead at higher levels of hierarchy as HALS will require shifting of lower level home regions more frequently than higher level homes. Unlike other location services, hierarchy generation includes two different grouping mechanisms i.e. 5x5 L0 cells create an L1 cell and after that 2x2 Li cells create an Li+1 cell as shown in Fig. 1. It should be noted that each Li cell must be member of exactly one Li+1 cell. Four Li-1 cells that compose Li are called sibling cells. The relationship between transmission range and size of L0 cells must be made in such a way that any two points in an L0 cell must be within communication range. As a result, if RTr is the transmission range and a is the width of an L0 cell then a = RTr Cos 45 0 = RTr 2 (1) B. Selection of Home Region Most of the grid based location services use a uniform hash function known to all nodes. These schemes may perform well in uniform node distribution but due to the mobility of nodes in a natural context where empty cells are frequent and node distribution is not uniform for long duration, significant amount of location updates and query failures are expected. This motivates us not to use hash function to map home region for the whole lifetime of HALS. For each L1 cell one of the L0 cells is assigned as Rank-0 (R0) Home, for each L2 cell one of the four R0 homes is assigned as an R1 home and this process continues for each level. A Rank-i home (Ri) is defined as an L0 cell, nodes within which act as server for nodes within its Li+1 cell. Let a node S working as a server in a cell with Ri where i ≥ 0 and another node A resides in its same Lj cell where 1 ≤ j ≤ i+1, S will store A’s location in the following format: ⎧ L cell id A resides in for j > 1⎫ Data( S , A) = ⎨ j −1 ⎬ ⎩ X A , Y A of A for j = 1 ⎭

(2)

An Li home for a node is defined as an Li cell in which its Ri home resides in. This implies an Ri home will act as R0 home for the nodes within the same L1 cell, act as R1 home for nodes within the same L2 cell and so on up to Ri home for nodes within the same Li+1 cell. So, an Ri home works as server of all nodes residing within its same Li+1 cell at different level of precision. As shown in Fig. 1 for any arbitrary node A, R0 is an L0 cell that acts as its L0 home. An L1 cell marked as H1 is L1 home of A. Within H1, R1 is actually working as server at Rank-1 for A. L2 cell marked as H2 is A’s L2 home. Within H2, R2 is Rank-2

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

home for A. It should be noted A does not have precise knowledge about the location of R2, it only knows that R2 is

Figure 1. Generating hierarchy and home region selection in HALS.

within the boundary of H2. As long as R2 is within H2, A does not need to update its L2 home information. Again, for another node B, R2 resides within its L1 cell, so R2 acts as B’s L0 home, the L1 cell within which R2 resides in is B’s L1 home and H2 is B’s L2 Home. R2 acts B’s R0, R1 and R2 home. So nodes in R2 know “in which cell nodes within the L3 cell boundary resides in” at different level of accuracy. Initially nodes are assumed to be distributed throughout the area, and every node is equipped with initialization function that will map each node with its home regions based upon its initial position at deployment. It should be noted that unlike hash function used by other methods, initialization function uses nodes initial deployment position rather ID and after initialization there is no use of this function any more. C. Home information dissipation Since HALS adopts home region shifting mechanism, no static hash function is applicable throughout the lifetime after deployment as home regions will change their position monitoring the node density in the neighboring cells. For this reason it is not possible to predict where the home region will be with respect to time. We employ home region information dissipation method that will help nodes to know the home regions at different level. HALS dissipates home information through HELLO packet. Since in GPSR each node sends beacons (HELLO) HALS includes following information with HELLO packet: a) home_id[]: an array of H elements where, H is the highest level in hierarchy, home_id[i] will contain the ID of Li home. b) Freshness: A time stamp that indicates how latest the home_id[] information is. c) no_of_nodes: number of nodes in HELLO generator’s L0 cell. Each node also stores these values in its database depending upon the freshness, and dissipates the most updated values it has. On receiving a HELLO from a neighbor, a node updates its database if FreshnessHELLO > Freshnessdb Only the nodes residing in Ri home where i ≥ 0 assign FreshnessHELLO with current time stamp, which implies the latest home info is actually generated from a Ranked cell and

other nodes only update their database and propagate it. This propagation is kept limited within the area of an L1 cell. So when an L0 home is shifted within an L1 cell, HELLO packets inform others the newly assigned home cell. If a node (at boundary of a cell) receives a HELLO from a node that does not reside in its same L1 cell, it does not consider updating the home information contained in that HELLO. The home_id values also adopt relative cell ID similar to how home servers store nodes location at different precision. For Li, i ≥ 0 home a node stores an Li cell ID in which its Ri home resides in. Thus even the Ri home is shifted as long as it remains within its Li cell boundary that node does not need to be notified. Since nodes propagate home_id through HELLO message, when a node crosses an Li boundary from cell c to c1 it receives HELLO from any node in c1 and updates its home info (when a node crosses Li cell boundary it needs to update its database for home value from level 0 to Li-1) , and starts propagating updated data. D. Forwarding packet to Rank-i home For location update, query and other message transfer a node needs to send packet to its home servers. If a node A is willing to send a packet to Rank-i home (Ri home) it only has the Li cell ID within which Ri home resides, but does not have the precise knowledge of Ri home’s position or ID. So A sends a packet with the following information towards the center of Li home of A: < Rank = i, Cellid of Li home of A > When the packet reaches to any node B inside destination Li cell, the Ri home for A actually works as Ri-1 home for B and resides inside the Li-1 home of B. So B modifies the tupple as: and forwards to the center of Li-1 home of B. In this way when the packet reaches within the boundary of an L1 cell where the Ri home resides, the packet is forwarded towards the destination, as every node within an L1 cell knows the location of its L0 home and any node within that L0 home can receive that packet. In this way a packet can be sent to a Rank-i home with Li home ID. It should be noted that forwarding in this way will select a non-optimal path from sender to destination. E. Shifting Home region To support non-uniform node distribution we adopt home region shifting mechanism. For each Ri home where i ≥ 0, a leader node is selected. Here we select a node as leader if it has the smallest ID within an Ri home. We color this leader as Red. When a red node moves away from its designated L0 cell, it marks itself as a normal node and sends a message to nodes within the old L0 cell so that a new red node can be elected depending upon the ID space. The red node is actually responsible for making decision of shifting home region. As every node sends the number of nodes present in the L0 cell it resides in with HELLO message, the red node stores this information and gets an assessment of how the node distribution within an L1 cell is changing. When it finds that the number of nodes in the current L0 home is less than a predefined value, it decides to shift home to any of the neighboring cells within the same L1 cell that has the highest number of nodes. We term this type of shifting as Soft shifting. The red node then handovers all of its server responsibilities to the nodes of the newly selected L0 home cell, sends a message

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

to elect a red node and finally marks itself as a normal node. Consequently newly selected L0 home starts attaching latest information about new home with HELLO and within a short time all nodes in the current L1 cell will update their database with newly changed data. It should be noted that propagating information with HELLO is not as fast as broadcasting but nodes do not need this updated information immediately. Since shifting home commences when there are still a few nodes in the old home they are capable of responding queries during this transition period. F. Destroying home region Soft shifting of Ri home cell is kept limited within the boundary of an L1 cell. Shifting home region will keep home region non-empty. If all nodes start to move away from an L1 cell, the Ri home will be the last to move, since, home region shifts to a higher density area. In an Li cell where i > 1, the highest ranked home works at least at Ri-1 which resides in one of the four Li-1 cells. If an Li-1 cell is going to be empty which does not contain the Ri-1 home, it informs the Ri-1 home that the corresponding Li-1 cell is going to be empty. In this way an Li home always know which of its siblings are empty. In case of the Li-1 cell where Ri-1 resides in is going to be empty, it selects one of the siblings within its Li cell to be responsible to continue its duty and increases the rank to Ri-1, transfers its database and destroy itself from acting as home and sends the Ri-2 home within the remaining two siblings that new home is updated. This type of shifting is termed as Hard shifting. Each of the ranked homes sends a message to each of its three siblings within the same Li-1 cell, and this process continues until Rank-0 home within it receives the information. It should be noted that, as long as a Ri home is within its Li cell, this message transfer is bounded within that Li cell. Consequently updated home information will be propagated through HELLO. G. Creating home region When a node crosses the boundary of an Li cell where i ≥ 1, it expects to get appropriate home information for residing in the new cell through HELLO packet originated from nodes already present there. It should be noted that when a node A crosses the boundary of a level i cell, from Cold to Cnew it also crosses all the level j cell boundary where j < i. In this case the following situations may happen: a) There exists nodes in Cnew from which A listens to HELLO and updates its database. b) A did not hear any HELLO within Cnew, because either there is no node in Cnew. So A does not have any idea about homes from L0 to Li-1. Since its Li home info is still up to date, it queries about Cnew whether Cnew is empty (while destroying a Ri home red node informs Ri+1 home). Upon receiving the reply A simply makes itself (the L0 cell) Ri-1 home and starts acting as server. c) Or, nodes are already there but not near enough for receiving HELLO. In this case A queries its Li home about the status of the Cnew. When it will get the reply that Cnew is not empty, it will start unicast query at fixed interval of time to the center of cells from L1 to Li-1 requesting about homes at L0 to Li-1 within Cnew. Since there exists nodes in Cnew, eventually the packet will be received by a node, and that node replies with the home information.

Depending upon the content of reply A can conclude about the missing home information and upgrades itself to the corresponding rank. H. Location Update Like other hierarchical location services HALS also updates its corresponding servers (homes) at different level of hierarchy with different level of accuracy. Location update can take place in one of the following ways: Time triggered, Distance triggered and Crossing cell triggered. When a node wants to send update its Li home it computes the Li cell ID in which it resides in. It knows the Li home ID which is available through HELLO. Therefore, the node sends its location in terms of Li cell ID to its Li home. The packet forwarding mechanism to Ri cell with Li cell ID is described in section D. In this way by means of cooperative forwarding packets can reach to the desired destination. In time triggered scheme, each node updates its location servers with fixed interval time triggered fashion. The interval for updating home at different level increases from lower level to upper level as lower level homes need to know more accurate location. Distance triggered scheme takes place when a node traverses a predefined distance. Crossing cell triggered scheme updates its Li home when it crosses Li cell boundary. It has the least overhead, but if update packet for higher level is lost for any reason, query failure may be substantially high. I.

Location Query If a node S wants to query the location of a target node T, the query packet needs to be routed to T’s R0 home, which contains the absolute position of T. The query is bounded by a smallest cell which contains both S and T. S sends the query to its current R0 home. If the home has T’s location it sends the reply. Otherwise it forwards the packet to its R1 home inside its L1 home. These forwarding continues until a home server detects T’s location. Let Lk , 0 ≤ m ≤ k home server has an entry of T—a cell ID of level m. The query is then forwarded to the highest ranked home within retrieved Lm cell. From that home the search area is reduced and an Lm-1 cell is retrieved where T resides in. In this fashion the packet reaches to a ranked cell as explained in Section D (that works as an R0 home for T) that contains the absolute location of T and replies to S with the location of T. J.

Scalability Issue Since HALS creates similar type of hierarchy development mechanism like HLS, the location update cost and query cost are similar to HLS having worst case complexity O(√N) in case of uniform node distribution where N denotes the number of nodes. But the storage requirement for HLS is O(logN) as each node works as server node for some other nodes. On the contrary, HALS location servers work more like centralized fashion, if the highest level in hierarchy is H then Rank-i servers have storage overhead of O(N/4H-i) and other nodes does not have any overhead at all. It should be noted in case of uniform node distribution, HALS home region shifting mechanism does not need to be activated, since in this case no cell is expected to be empty. On the contrary if node distribution is non-uniform, the performance of the existing hash function based location services degrades substantially while HALS overcomes this problem with shift, create and destroy mechanisms of home regions at the expense of some

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

IV.

SIMULATION RESULTS

For implementing our proposed location service HALS we used Network Simulator NS 2.29 and its performance was compared to HLS [9], because its implementation is available at http://www.cn.uni-duesseldorf.de/staff/kiess/software/ and similar to HIGH-GRADE from design point of view and balances update and query cost while being highly scalable. TABLE I presents the values of basic parameters used for comparing the two location services and most of these values were taken from [9] as well. We selected Random Waypoint (RWP) model as mobility model of nodes with 1 sec pause time for analyzing uniform node distribution. Formation of Region of Interest (ROI) will results in higher number of nodes concentrated in specific regions enabling us to simulate how proposed and existing schemes behave in such situation. Therefore, to investigate how the two methods perform in nonuniform node distribution we adopted random waypoint mobility model with ROI. 2 rectangular shaped ROIs of 600x400m and 500x500m were selected in upper left and lower right portion of the simulation environment. In presence of ROIs a node randomly selects one region and within it a point is selected randomly and tends to move there. After arriving there it pauses for 1 sec and again selects a region and this process continues until the simulation time expires. This type of mobility is observed in presence of hot spots where people are only interested in visiting those spots. Various parameters related to performance evaluation of location service protocols, including location update and query success rate at a wide range of node movement speed, node density and transmission range were calculated. For space limitation results for 200m transmission range is presented here. Similar trend in result is also found for 250m range. TABLE I.

PARAMETERS FOR SIMULATION

Number of nodes N Area size Transmission range Rtr Max node speed Mobility model Simulation time Request per node Number of runs MAC layer Routing Protocol

200, 300 and 400 2000m X 2000m 200m and 250m 5, 10, 15 and 20 m/sec Random waypoint model with & without ROI 300 sec 4 10 IEEE 802.11 GPSR

Update success rate (%)

100 95 90 HALS 85

HLS

80 75 70 0

5

10

Max speed (m/sec)

15

20

Figure 2. Location update success rate at uniform node distribution.

Fig. 2 shows the location update success rate of HLS and HALS at different speed at 400 nodes in random waypoint model. HALS shows higher update success rate over HLS

because the home region shifting mechanism never lets the home region to become empty even for a short duration of time. Moreover, HALS sends update packets in both greedy and perimeter mode unlike HLS which uses greedy mode only. Though random waypoint model tends to scatter the nodes throughout the simulation area, some of the cells become empty (for short duration) and during this period no update packet can be received successfully. For this reason though in HLS there is a mechanism to store and forward pending location information to the designated server cell in case of being empty, due to node mobility this scheme does not guarantee successful update as observed in Fig. 2. 100

Query success rate (%)

overhead. By selecting higher ranked cells in an intelligent way e.g. nearby hot spot regions, number of shifting can be reduced and can improve the overall performance.

95

90 85 HALS

80

HLS 75 0

5

10

15

20

Max speed (m/sec)

Figure 3. Location query success rate at uniform node distribution.

Fig. 3 shows the query success rate of HALS and HLS with at different speed in random waypoint mobility model with 400 nodes. The result shows HALS performs better than HLS. As the node movement speed increases the performance of HLS degrades more rapidly than HALS. Every node in HLS is involved in handover of location information when they cross the smallest cell. As a result at increased speed the rate of handover increases as well. Where as in HALS the number of home cells are less than that of HLS, the handover process is limited in ranked cells only. So, increased node mobility has less affect on HALS. Moreover, higher number of cells for composition of L1 cell results in less hierarchy levels in HALS. Furthermore, lower update success rate causes relatively lower query success rate because of failure in update servers properly. In case the location is not updated properly, the servers can not provide up to date location. The interesting part is though the location update success rate in HLS is significantly lower, the query success rate is relatively higher. In the implementation of HLS we found that the location caching mechanism has been implemented which improves the query success rate. While a node forwards an update packet towards destination, all the intermediate nodes engaged in forwarding the packet and other nodes that can listen to the packet forwarding in MAC layer by overhearing also store the location information. Thus it can be concluded that a high amount of location query is answered not from its designated home regions, rather from other nodes that have cached data and query success rate is relatively higher compared to update success rate. On the other hand since the success rate of location update rate in HALS is higher and home regions never become empty, the query success rate also becomes higher than HLS. Fig. 4 shows update success rate at different speed when ROIs are present in case of 400 nodes. When the ROIs are present nodes tend to travel towards ROIs in the beginning and then from one region to another. This makes node movement much concentrated on the ROIs and interconnection of ROIs leaving almost no nodes available in other portion of area. This

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

causes a vast amount of cells to be empty which in turn decreases location update success rate of HLS. At higher node movement speed cells become empty at a higher rate and that causes the update success rate drop dramatically at higher speed. But for HALS the drop in update success rate is less sensitive to node mobility than HLS. The effect of lower update success rate has direct impact on the measurement of query success rate of HLS which drops drastically at higher node speed which is shown in Fig. 5. As observed in the Fig. 5 at 10m/sec speed HALS has 92% success rate compared to 85.8% of HLS, but when the speed is increases to 20m/sec, HALS shows 87.5% compared to 71.5% of HLS. 100

Update success rate (%)

95 90 85 80 75 70 65 60

HALS

55

HLS

50 0

5

10

Max speed (m/sec)

15

20

Figure 4. Location update success rate at non-uniform node distribution. 100

Query success rate (%)

95

b) Node speed and density has less impact on update and query success in HALS than HLS. c) Though HALS shows superior performance in terms of location update and query in both uniform and nonuniform node distributions, storage overhead for servers is not uniformly distributed among nodes. V.

Existing location service protocols are designed assuming uniformly distributed nodes. Therefore, their performance degrades significantly for non-uniform node distribution which may arise in many practical scenarios where nodes may concentrate in specific regions at a particular time and at higher mobility of nodes. To address this problem we propose a hierarchical location service that dynamically creates, shifts and destroys home regions making it capable of coping well with empty home regions resulting from non-uniform node distribution while maintaining scalability. The characteristics of this location service are analyzed in details with varying node density and distribution and simulation results show significant improvement in location service measurement metrics. Further study to employ an analytical model and using a wide range of scenarios in presence of obstacles and realistic node movement pattern [14] is needed to comprehend on the overall performance of proposed location service and remains the future focus of our study.

90

REFERENCES

85

[1]

80 HALS

75

HLS

70

[2]

65 0

5

10

Max speed (m/sec)

15

20

Figure 5. Location query success rate at non-uniform node distribution. HALS(RWP)

HALS(ROI)

HLS(RWP)

HLS(ROI)

[3] [4] [5]

95

Query success rate (%)

CONCLUSION AND FUTURE WORKS

90

[6]

85 80

[7]

75 70 65 150

200

250

No of nodes

300

350

400

Figure 6. Location query success rate at different node density.

Fig. 6 shows the query success rate at different node density with and without ROIs at 20m/sec. HALS shows similar trend in graph in performance with and without ROIs, which implies that it is less sensitive for varying node distribution. On the other hand the non-uniform node distribution makes significant difference in query success rate for HLS compared to uniform distribution. In random waypoint as node density increases from 200 to 400 the query success rate increases, but in presence of ROI the query response becomes more or less insensitive. From the above analysis, we conclude: a) By not allowing home region to be empty, HALS increases update and query success rate.

[8]

[9]

[10] [11]

[12] [13] [14]

P. Bose, P. Morin, I. Stojmenovic, and J. Urrutia, "Routing with guaranteed delivery in wireless ad hoc networks," Wireless Networks, vol. 7, pp. 609-616, 2001. Y.-B. Ko and N. H. Vaidya, "Location - Aided Routing (LAR) in Mobile Ad Hoc Networks," in MOBICOM, USA, 1998, pp. 66-75. B. Karp and H. T. Kung, "GPSR: Greedy perimeter stateless routing for wireless networks," in MOBICOM, USA, 2000, pp. 243-254. S.-C. M. Woo and S. Singh, "Scalable Routing Protocol for Ad Hoc Networks," Wireless Networks, vol. 7, pp. 513-529, 2001. C. Cheng, H. Lemberg, S. Philip, E. v. d. Berg, and T. Zhang, "SLALoM: A Scalable Location Management Scheme for Large Mobile Ad-hoc Networks," in WCNC, 2002, pp. 574-578. Y. Xue, B. Li, and K. Nahrstedt, "A scalable location management scheme in mobile ad-hoc networks," in LCN, USA, 2001, pp. 102-11. B.-C. Seet, Y. Pan, W.-J. Hsu, and C.-T. Lau, "Multi-home region location service for wireless ad hoc networks: an adaptive demanddriven approach," in WONS, 2005, pp. 258- 263. J. Li, J. Jannotti, D. S. J. D. Couto, D. R. Karger, and R. Morris, "A Scalable Location Service for Geographic Ah Hoc Routing," in MOBICOM, 2000, pp. 120-130. W. Kiess, H. Fubler, J. Widmer, and M. Mauve, "Hierarchical location service for mobile ad-hoc networks," Mobile Computing and Communications Review, vol. 8, pp. 47-58, 2004. Y. Yu, G.-H. Lu, and Z.-L. Zhang, "Enhancing Location Service Scalability with HIGH-GRADE," in MASS, 2004, pp. 164-173. I. Stojmenovic, D. Liu, and X. Jia, "A Scalable Quorum-based Location Service in Ad Hoc and Sensor Networks," Communication Networks and Distributed System, vol. 1, pp. 71-94, 2008. J.-R. Jiang and W.-J. Ling, "SEEKER: An Adaptive and Scalable Location Service for Mobile Ad Hoc Networks," in ICS, 2006. I. Abraham, D. Dolev, and D. Malkhi, "LLS: A locality aware location service for mobile ad hoc networks," in DIALM-POMC, 2004, pp.75-84. S. Ahmed, G. C. Karmakar, and J. Kamruzzaman, "Geographic Constraint Mobility Model for Ad Hoc Network," in MASCOTS, USA, 2008, pp. 337-346.