service location protocols for mobile wireless ad …...service location protocols for mobile...

Int. J. Ad Hoc and Ubiquitous Computing, Vol. 2, No. 1, 2006 1

Copyright © 2004 Inderscience Enterprises Ltd.

Service Location Protocols for Mobile Wireless Ad Hoc Networks C. K. Toh

1), Guillermo Guichal

2), Dongkyun Kim

3), and Victor O.K. Li

4) Department of Electronic Engineering, Queen Mary, University of London, UK

1) Embedded Systems Ltd, Argentina

2) Department of Computer Engineering, Kyungpook National University, Korea

3) Department of Electrical and Electronic Engineering, University of Hong Kong, China

4) Email: [email protected]

1), [email protected] 3), [email protected]

4)

Abstract: As service-oriented networks are essential in our lives, service location, access and provisioning over wired or wireless networks govern users’ satisfaction. While they are well understood over cellular networks, it is not clear for ad hoc networks. We therefore present several approaches on service location in such networks: distributed broadcast query, query via service coordinators, and a hybrid combination. Particularly, we evaluated these approaches in terms of service availability and control overhead over variable network sizes and mobility scenarios. Simulation results showed that the distributed query approach suffered from high control overhead despite its robustness and tolerance towards mobility. The service coordinator (SC) based approach, however, resulted in a more scalable architecture that tends to separate the network into different service areas. This has important implications on future “mobile marketplace” and services to mobile users. Finally, the hybrid approach provided greater flexibility and robustness to the SC-based approach with its scoped broadcast to limit the overhead.

Keywords: Service discovery protocols; ad hoc networks; distributed broadcast; service coordinator; service availability; control overhead.

Biographical notes: C.K. Toh is a Professor with the University of London Queen Mary. Previously, he was the Director of Research, Communication Systems, Northrop Grumman Corporation. Earlier on, he was on the faculty at Georgia Institute of Technology (USA). He had authored two pioneering books: "Wireless ATM & Ad Hoc Networks" (Kluwer Academic Press, 1996) and "Ad Hoc Mobile Wireless Networks" (Prentice Hall Engineering Title Best Seller). IEEE President named him the recipient of the 2005 IEEE Institution Kiyo Tomiyasu Medal. He was the technical editor for several IEEE, ACM and KICS journals and was a member of IEEE Meetings and Conferences Board and a steering committee member for IEEE Transactions on Mobile Computing. He is a Fellow of the British Computer Society, the IEE, the New Zealand Computer Society, the Hong Kong Institution of Engineers, the Cambridge Commonwealth Society, and the Cambridge Philosophical Society.

Guillermo Guichal received his MSEE degree from Georgia Institute of Technology, USA in 2001. He is now with Embedded Systems Ltd in Argentina. His research interests are on wireless networks, computer protocols and the internet.

Dongkyun Kim is a professor in the Department of Computer Engineering, Kyungpook National University, Daegu, Korea. He received the BS degree at Kyungpook National University. He also obtained the MS and Ph.D degrees at Seoul National University, Korea. He was a visiting researcher at Georgia Institute of Technology. He also performed a post-doctorate program at University of California Santa Cruz. He has been a TPC member of several IEEE conferences. He received the best paper award from the Korean Federation of Science and Technology Societies, 2002.

Victor O.K. Li received SB, SM, EE and ScD degrees in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology, Cambridge, Massachusetts, in 1977, 1979, 1980, and 1981, respectively. He joined the University of Southern California (USC), Los Angeles, California, USA in February 1981, and became Professor of Electrical Engineering and Director of the USC Communication Sciences Institute. Since September 1997 he has been with the University of Hong Kong, Hong

2 C.K. TOH, GUILLERMO GUICHAL, DONGKYUN KIM, AND VICTOR O.K. LI

Kong, where he is Chair Professor of Information Engineering at the Department of Electrical and Electronic Engineering. He has also served as Managing Director of Versitech Ltd, the technology transfer and commercial arm of the University, and on various corporate boards.

1 INTRODUCTION

The success of cellular communication service has brought about substantial growth in the mobile phone business over the years. Mobile cellular communications can now carry not only voice but also useful data, including Internet and web-browser traffic. The network architecture today has three primary sections, namely: (a) access network, (b) core network, and (c) Internet. With such an architecture, it reflects the heterogeneity in the network setup. With innovations in short-range mobile communications, Bluetooth[5] and peer-to-peer computing have evolved. Bluetooth networks represent a new generation of mobile networks that are infrastructureless, self-organizing, and comprise all mobile elements. Supporting this Bluetooth network is the presence of a multi-hop routing protocol, a master node arbitrated media access method, and the use of time-switched frequency hopping mechanism.

While Bluetooth networks support both piconet and scatternet communications, it has yet to support communications beyond scatternets. Ad hoc mobile networks can be considered as extensions of Bluetooth, where multihop communications can occur beyond scatternets. Hence, in ad hoc mobile networks, issues related to routing (both unicast and multicast) and channel access present plenty of scope for early research in this area.

As the underlying communication issues in ad hoc mobile networks are gradually investigated and resolved, thoughts about potential applications and service usage scenarios have evolved. Since ad hoc networks are self-organizing networks and can be deployed quickly without intervention from operators and system administrators, it is hard to envisage what the service model would be like. However, prior to thinking about service provisioning in ad hoc networks, one should also need to address service location. Without an accurate, efficient, fast, and robust service location mechanism, service access and provisioning would be difficult and problematic.

Current service location protocols proposed to the IETF are primarily aimed at wired networks. For example, service location protocol (SLP)[1] is an open standard specification that enables clients to locate services in campus-wide IP networks. It does not address mobility of service agent, directory agents, and client agents. Universal plug and play (UPnP)[3], on the other hand, addresses the problem of auto-configuration by using DHCP and DNS. However, in ad hoc networks, it is difficult to have a fixed DNS since the DNS node can also move about. In [6], community-based service location is proposed for the wired internet, rather than ad hoc wireless.

Jini[2] is one commercial proposal that uses the concept of federation, where a group of devices can register with each other to share services. Each Jini subsystem contains a set of

lookup services that maintain dynamic information about services in the network and the location of these services can be known in advance or discovered using multicast. However, Jini does not consider broken communication links due to mobility, i.e., it cannot function well when host communications are intermittent.

Bluetooth (BT)[5] is a wireless network that has close resemblance to ad hoc networks except that it is structured in the form of piconets and scatternets. The service discovery protocol proposed in Bluetooth allows a device to discover resources by polling the other devices in the piconet. A SDP server holds information about services present in a device and the services’ properties. BT-SDP does not define how services are accessed but only allows clients to find these services and read their attributes. In addition, BT-SDP is not applicable to generic multi-hop wireless networks due to its limitations with scatternets.

1.1 Design Considerations

Prior to designing or proposing specific service location protocols for ad hoc wireless networks, the following factors need to be considered:

a. Ability for a service to announce itself and its capabilities or properties to the network

b. Automatic discovery of services available in the network or within a limited portion of the network

c. Self-configuration of services and clients without administrative intervention

d. Allow devices to recognize the protocols needed to access a specific service, and allow seamless interoperability whenever possible

In addition to the above functional considerations, an ad hoc service location protocol should also address the following:

a. Low communication bandwidth – Wireless communication capacity has yet to match that of wired networks and is limited by noise, interference, contention, etc. A well-designed protocol can help to improve performance by reducing control overhead and using available bandwidth wisely.

b. Mobility – Devices are mobile and hence network topology is dynamically changing with time. Links are broken and new connections are formed during mobility. The service location protocol must be able to operate during such an environment.

c. Power constraints – Most mobile devices are powered by batteries and hence have limited period of operation prior to the next charge cycle. Hence, a service location protocol should not excessively consume power.

SERVICE LOCATION PROTOCOLS FOR MOBILE WIRELESS AD HOC NETWORKS 3

This paper is organized as follows. Section 1 discusses limitations of current service location methods and the motivation for new schemes in ad hoc mobile networks. Section 2 describes the three methods proposed, namely: (a) distributed query, (b) SC-based, and (c) hybrid service location. Section 3 presents the simulation environment, parameters and discusses the results on service availability and control overhead for the distributed query method, SCs-based and hybrid schemes. Section 4 presents an overall comparison for all schemes, under varying hop counts and percentage of servers. Finally, a conclusion is provided in Section 5.

2 PROPOSED SERVICE LOCATION SCHEMES

2.1 Distributed Query Approach

The distributed query approach does not rely on any centralized repository for service information or network-wide server lookup. As shown in Figure 1 below, when a client requires a service (i.e., access the service to a server), it queries the network via a limited broadcast message. Nodes receiving the query packet will process it and decide whether to propagate it further. Nodes that have previously forwarded a query packet will not rebroadcast the same query packet again. Servers that can provide requested services will reply. The client then selects the most convenient or adequate server to access desired services. In this paper, our discussion is based on mobile devices that use radios with omni-directional antennas.

Figure 1 Query Service Location Scheme – (a) Clients broadcast their service query, and (b) available servers

respond to specific clients. The lookup service in the client host can cache the replied server information to avoid new queries for the same service in the future. Such cache entries should have an expiration time after which they are assumed to be stale (for example, the server previously discovered has now moved away and

is out-of-reach) and should be deleted. Figure 1 illustrates the client query broadcast and server unicast reply approach in this scheme.

2.2 SC-based Approach

In the service coordinator (SC) based approach, a number of special nodes known as service coordinators, are chosen to store servers and services information and act as brokers on behalf of clients.

As shown in Figure 2, SCs advertise their presence to neighboring nodes. Servers then register their services with the neighboring SC, and the SC keeps a table of available services in its area of influence.

Clients needing a service will query their neighboring SCs to see if they have entries that match the desire service. If so, the appropriate SCs will respond with information on which servers the clients can contact and negotiate directly for services. If multiple servers can fulfill desired services, the decision on a specific server to use is left to the client.

To enable scalability, SCs will influence a limited region of the network. This region can be called a service cluster, and can be varied dynamically by changing the number of hops the SC chooses to propagate its advertisements. To prevent excessive overhead, SCs will not forward another SC’s advertisement unless it will reach nodes farther than its own. As servers learn about SCs from the advertisements, servers register their services with the SCs.

To allow for greater service coverage, SCs may forward registration messages to other SCs they are aware of. When clients in an SC’s service cluster need a particular service, they can query the SC. Clients select an SC based on the SC advertisements that they receive. If multiple SCs’ advertisements are heard, then the clients can decide on a particular SC to send their service queries to.

Figure 2 Service Coordinator Scheme – (a) SCs advertise themselves, (b) servers register with SC, (c) clients query SC according to their chosen SC, and (d) clients access

servers.


2.3 Hybrid Approach

In the SC-based scheme, if there is no SC available to a given client, then it is unable to access services. This limitation can be overcome in the hybrid scheme.

The hybrid scheme basically combines the features of the query and SC-based schemes. In this scheme, there shall be SCs, servers, and clients in the ad hoc service provisioned network. If a client is aware of a SC based on the received advertisements, it will query that SC for service information. However, if there are no SCs available, it will fall back to the query approach to search for a possible service provider/server. This has the added advantage that it allows nodes in areas where there are no devices that can or want to act as SCs to access services around them. Figure 3 illustrates how this scheme operates.

Figure 3 Hybrid Scheme – (a) SC advertises to neighbors, (b) servers register with SC, (c) client sends query in search of servers, and (d) servers respond to client’s request, not

the SCs in this scenario.

3 SIMULATION AND RESULTS

3.1 Simulation Environment

To evaluate the performance of the proposed schemes, simulations are carried out. These three schemes are implemented into our discrete-event simulator and nodes are randomly positioned on the 80 x 80 m grid. Nodes have a communication range of 10m. The simulation parameters used are presented in Table 1.

The service location schemes are evaluated under both static and mobile scenarios. For the static case, a SC advertisement and service registration is initiated, and clients perform only one service request. For the mobile case, nodes migrate using a random move-and-stop model. Figure 4 shows two snapshots of the resultant ad hoc network topology after migration over time. Note that at any instance in time, the network is never partitioned and this is ensured in our simulator for the purpose of this evaluation.

Table 1 Simulation Parameters for Static and Mobile Scenarios

Parameter Static Scenario Mobile Scenario Number of nodes (including clients, servers, and/or service coordinators)

50 50

Percentage of clients 30% 30% Percentage of servers 5, 10, 20 and 30% 5, 10, 20, and 30%Percentage of service coordinators

0, 5, 10, 20, 30, and 40%

0, 5, 10, 20, 30, and 40%

Grid size 80 x 80 m 80 x 80 m Communication range 10m 10m Speed - 2 – 3 m/s Probability of moving - 80% (allowing for

some rest/pause periods)

Simulation time Instant 100 seconds Broadcast or advertising hops

1, 2, and 3 1, 2, and 3

Number of simulation runs

500 100

Service request arrivals 1 per client Poisson process, mean of 1 second

Network average degree of connectivity

3.2 2.3

Figure 4 Snapshots of ad hoc wireless network topologies. The two performance parameters of interest are: (a) service

availability, and (b) control overhead. Service availability gives an indication on how frequently service requests can be fulfilled. From a service provider’s point of view, this is important since it directly impacts service provisioning.


From a customer’s point of view, this factor determines how useful his device is in terms of accessing for desired services. The control overhead is another important factor since deploying a scheme that incurs too much overhead will render inefficient use of available network capacity. Simulation results related to these two parameters are gathered under varying percentage of servers, service coordinators, and radio hop counts. Detailed evaluation of these results is presented in the next section. Note that in our simulation, the query-based scheme does

not implement caching of queried information. Instead, clients query the network every time they require a service. Hence, we evaluate performance of the query method without using caching.

3.2 Evaluation of Location Query Method

Figure 5 Query Scheme Service Availability Performance – (a) static scenario, and (b) mobile scenarios with varying

hop counts and percentage of servers.

As shown in Figure 5, more servers generally result in an increase in service availability. Also, more hops yield more

availability because more nodes are accessible for services. When mobility is introduced, service availability improves when the percentage of servers is low. This is because mobility increases the chance of a client “meeting” a server. When the percentage of servers is large, this effect becomes marginal.

Figure 6a reveals that control overhead increases with increasing hop count and increasing percentage of servers. However, increasing servers has slight impact on control overhead at 1 hop since only servers 1-hop away from the requesting client will respond, contributing marginal increase in overhead. At large hops and with more servers, more servers will reply, explaining the observable increase in overhead. Under the presence of mobility, control overhead is significantly increased, as shown in Figure 6b. This is due to retransmissions of service requests. The results seem to imply that the query method is not efficient and scalable when the network size and the percentage of servers are large.

Figure 6 Query Scheme Control Overhead Performance – (a) static scenario, and (b) mobility scenarios with varying

hop counts and percentage of servers.

Availability for Distributed MethodStatic Tests

0.0000.100

0.2000.300

0.4000.500

0.6000.700

0.800

0.900

1.000

5 10 15 20 25 30

% Servers

Ave

rage

ava

ilabi

lity

1hop queries

2 hop queries

3hop queries

Availability for Distributed MethodMobile Tests

0.000

0.100

0.200

0.300

0.400

0.500

0.600

0.700

0.800

0.900

1.000

5 10 15 20 25 30

% Servers

Avg

ava

ilabi

lity

1 hop queries

2 hop queries

3 hop queries

Overhead for Distributed MethodStatic Tests

0

50

100

150

200

250

300

5 10 15 20 25 30

% Servers

Tota

l msg

s 1hop queries

2 hop queries

3hop queries

Overhead for Distributed MethodMobile Tests

0

2000

4000

6000

8000

10000

12000

5 10 15 20 25 30

% Servers

Tota

l Mes

sage

Ove

rhea

d

1 hop queries

2 hop queries

3 hop queries


3.3 Evaluation of SC-based Method – Service Availability

3.3.1 One-Hop Case – Service Availability

In the SC-based scheme, we measure service availability and overhead with varying percentage of servers and service-coordinators. Results plotted in 3-dimensional graphs are shown in Figures 7-12. For the 1-hop case, an increase in SCs results only in marginal increase in service availability (Figure 7a). However, if the percentage of servers is concurrently increased, we observe significant increase in service availability. This result indicates that for low percentage of servers, populating the network with more SCs does not improve service availability – a particularly important observation for potential ad hoc wireless service providers. From Figure 7b, service availability is reduced with the presence of mobility. Significant reduction is observed at high percentage of SCs and servers. Hence, mobility of SCs and servers has the effect of negating the ability to access services.

Figure 7 Service-Coordinator Scheme Service Availability Performance – (a) 1-hop static, and (b) mobility scenarios

with varying percentage of servers and SCs.

3.3.2 Two-Hop Case – Service Availability

Compared to the one-hop static case, service availability has increased substantially with two-hop scenarios, as shown in Figure 8. At 30% servers and 40% SCs, service availability has increased from 59% to 92%, as shown in Figures 7a and 8a. This indicates that more services can be discovered when the search radius is expanded.

Generally, with increasing SCs and percentage of servers, service availability increases. The gray zones on the graph in Figure 8a depicts that if a service provider wishes to provide 60-70 percent availability, he can operate with 20-35 percent of SCs and 15-20 percent of servers. Again, with the introduction of mobility, service availability is reduced, as indicated by the relatively tapered contour shown in Figure 8b.

Figure 8 Service Coordinator Scheme Service Availability Performance – (a) 2-hop static, and (b) mobility scenarios


3.3.3 Three-Hop Case – Service Availability

Service availability increases significantly with 3 hops, especially when the number of SCs exceeds 10%. Also, it is


clear that beyond 10% servers and at high percentage SCs, service availability gradually reaches saturation. Compared to the 2-hop case, service availability has increased from 40% to 60% at 40% SCs and 5% servers.

Under the presence of mobility at 3-hop case, service availability has been lowered, as shown in Figure 9b. Compared to the 1-hop and 2-hop cases, service availability is still higher in the 3-hop case. Note also the changes in the service availability graphs for 1, 2 and 3 hops. The graph contour basically gets inflated with increasing hop counts, reaching saturation at some points.

Figure 9 Service-Coordinator Scheme Service Availability Performance – (a) 3-hop static, and (b) mobility scenarios


3.4 Evaluation of SC-based Method – Control Overhead

3.4.1 One-Hop Case – Control Overhead

The amount of control overhead increases with increasing percentage of SCs and servers. Hence, one should avoid introducing a service location protocol with an overwhelming number of servers and SCs, which will result in significantly lower capacity for data transfer.

For static case, overhead increases with increasing SCs and servers. As shown in Figure 10a, at 10% or more SCs and with 30% servers, control overhead increases significantly. This has important implications on populating an ad hoc network with the appropriate number of servers and SCs. Unlike the static case, a significant amount of overhead occurs in the network when there is mobility. As shown in Figure 10b, more control packets are transmitted, about 100 times more than the static case. Hence, for a service framework on an ad hoc mobile network under the presence of mobility, incremental increases in SCs and servers could result in explosive growth in control overhead, resulting in less data delivery to users.

Figure 10 Service Coordinator Scheme Control Overhead Performance – (a) 1-hop static, and (b) mobility scenarios


3.4.2 Two-Hop Case – Control Overhead

With 2 hops, more control overhead is incurred. Compared to 1-hop static case, the maximum number of control packets is now 350 instead of 128, as shown in Figures 10a and 11a. Note that at 5% servers and 40% SCs, control overhead is at 120 messages but this increases to 7648 messages when mobility is present. With the presence of


mobility, the worse case overhead has increased from 350 to 19213 control packets.

When compared to 1-hop with mobility, control overhead has increased in the 2-hop case. At 5% servers and 40% SCs, control overhead has increased from 4025 to 7648 messages. At 40% SCs and 30% servers, worse case overhead has increased from 10570 to 19213 messages, as shown in Figures 10b and 11b.



3.4.3 Three-Hop Case – Control Overhead

At 3 hops with no mobility, overhead is further increased. At 5% servers and 40% SCs, control overhead is at 200 packets, an increase of 66.67% when compared to the 2 hop static case (it was 120 packets). The maximum control overhead is now at 696 packets instead of 350 packets in the 2 hop static case, an increase of 98.86%. With mobility, at 5% servers and 40% SCs, control overhead has increased from 200 packets to 10153 packets, as shown in Figures 12a and b, respectively. The maximum control overhead generated was above 24000 packets, compared to 696 in the static case.

All these seem to indicate that increasing hop count, %SCs, %servers, and the introduction of mobility has a profound effect, resulting in significantly large control overhead.



3.5 Evaluation of Hybrid Location Method – Service Availability

3.5.1 One-Hop Hybrid Case – Service Availability

In the hybrid scheme, SCs advertise themselves within one radio hop and clients performed query search also within one radio hop. Compared to SC-based service availability results, the graph now exhibits a different contour. Service availability increases almost linearly with increasing percentage of servers for 1-hop broadcast. In addition, increasing the percentage of SCs only has a marginal effect on service availability in the hybrid scheme. This is a significantly different behavior when compared to the results in the SC-based location method described previously. With the presence of mobility, service


availability has improved at lower percentage of servers and SCs. This is so because with mobility of servers, SCs, and clients, more clients managed to “meet” and get access to servers. At high percentage of servers and SCs, mobility has little effect on service availability, which explains similar results for both static and mobility case, as shown in Figures 13a and b.

Figure 13 Hybrid Scheme Service Availability Performance – (a) 1-hop static, and (b) mobility scenarios


3.5.2 Two-Hop Hybrid Case – Service Availability

With 2-hop broadcast in the hybrid scheme, service availability has improved. At 40% SCs and 5% servers, service availability is 41.6%, compared to 14.7% in the 1-hop case. At 30% servers, increasing SCs beyond 5% only has marginal effect on service availability, as shown in Figure 14a. With the presence of mobility, service availability has improved for the scenarios with low percentage of servers. At high percentage of servers and SCs, service availability becomes close to the static case.



3.5.3 Three-Hop Hybrid Case – Service Availability

With the increase in radio range, service availability has further improved, as shown in Figure 15a. The graph contour is now flatter and 96% service availability is reached with 20% servers and 5% SCs or more. At 5% servers and 5% SCs, service availability is now at 52%, compared to 31% in the 2-hop case and 11% in the 1-hop case. This seems to indicate that adding an addition hop of SC advertisement and client query capability increases service availability by approximately 20%.

With the presence of mobility, service availability has further improved in the 3-hop scenario. Mobility of SCs, servers, and clients here has the tendency to allow higher success in accessing a reachable server. At 40% SCs and with 5% servers, service availability is at 72%, compared to 61% for the static case. Clearly, throughout the simulation time, mobility of nodes did result in improvement in overall service availability in the hybrid scheme.




3.6 Evaluation of Hybrid Location Method – Control Overhead

3.6.1 One-Hop Hybrid Case – Control Overhead

The hybrid location method yields slightly higher overhead performance as in the SC-based scheme. At 5% SC and 5% servers, control overhead equals 24 messages, compared to 11 messages in SC-based static case. At the other extreme, i.e., 30% servers and 40% SCs, 133 control messages are generated, compared to 128 messages in SC-based static case (an increase of 4.7%).

Under the presence of mobility, control overhead has increased considerably. At 5% servers and 40% SCs, 4857 control messages were generated, compared to only 51 messages in the static case. This is an increase by 95 times. At 30% servers and 40% SCs, 11205 messages have occurred; compared to 133 messages in the static case (this represents an increase of 84 times). However, in the 1-hop mobility SC-based case, control overhead was at 10570 messages, which is an increase of 6%. Hence, it is clear that mobility increases control overhead significantly, when compared to the static case.

Figure 16 Hybrid Scheme Control Overhead Performance – (a) 1-hop static, and (b) mobility scenario with varying

percentage servers and SCs.

Table 2 Control Overhead Comparison – 1 Hop Case

Observation At 5% SC, 5% servers, with mobility

At 40% SCs,5% servers

At 30% servers, 40% SCs

Remarks

SC-based Case

11 messages (static)

1323 messages (mobility)





Mobility increases control overhead significantly

Hybrid Case 24 messages (static)


4857 messages (mobility); 51 messages (static)



Mobility increases control overhead significantly

Note that the hybrid scheme yields the highest control overhead among all other schemes considered under the presence of mobility. This is because similar to the SC-based scheme, overhead was incurred with advertisements and an additional overhead of location queries was made if the hybrid scheme reverts to broadcast query when the SC-


based search failed. Note that this remark applies to cases with higher hop counts. Table 2 summarizes our findings.

3.6.2 Two-Hop Hybrid Case – Control Overhead

With the increase in radio hop count, control messages have also increased. At the upper extreme, i.e., 30% SCs and 40% servers, 352 messages were generated, compared to 133 messages in the 1-hop static case. At 5% servers and 5% SCs, 74 messages were generated compared to 24 messages in the 1-hop static case. With the presence of mobility, control overhead is further increased. At 30% servers and 40% SCs, 20165 control messages exist compared to 11205 messages in the 1-hop mobility case. At 5% servers and 5% SCs, 6432 messages were generated, compared to 3197 messages. This clearly shows that increasing the hop count increases control overhead.



3.6.3 Three-Hop Hybrid Case – Control Overhead

At 3 hops, control overhead has magnified, especially at higher percentage of SCs and servers. The following discussion shall refer to Figures 17a and 18a. At 40% SCs

and 30% servers, 696 control messages were generated, compared to 352 messages in the 2-hop static case. At 40% SCs and 5% servers, 200 control messages were generated, compared to 122 messages in the 2-hop static case. At 5% SCs and 5% servers, 113 control messages are present compared to 74 messages in the 2-hop static case. With the presence of mobility, control messages are further increased. While 20165 control messages are observed at 40% SCs and 30% servers for the 2-hop mobility case, it is now 27211 messages. At 40% SCs, considerable increase in overhead is observed beyond 10% servers. At 5% SCs and 5% servers, 8673 control messages exist compared to 6432 messages in the 2-hop case. Hence, increasing hop count and mobility increases control overhead in the hybrid scheme.



4 SUMMARY & OVERALL COMPARISONS

4.1 Discussion on Service Availability of all 3 Methods


Table 3 Observations on Service Availability

With increasing percentage SCs

With increasing percentage of servers

With increasing hop count

With presence of mobility

Query case

Not Applicable

Service availability increases

Service availability increases

Minute changes at high percentage of servers. Improved service availability at low percentage of servers

SC-based case

Service availability generally increases

Service availability only increases significantly when % SCs is large

Service availability increases significantly

Mobility generally reduces service availability at high % of SCs and servers but improves service availability at low % of servers and high % of SCs

Hybrid case

Marginal effect on service availability

Significant increase in service availability

Service availability increases significantly

Mobility reduces service availability marginally at high % of SCs and servers but improves service availability at low % of servers across different percentage of SCs.

As shown in Table 3 above, we summarize the characteristics of the three proposed schemes, under the four varying conditions covered in our work. With increasing SCs, service availability is generally increased in the SC-based scheme, but not significantly in the hybrid scheme, as shown by Figures 7a, 8a, 9a and Figures 13a, 14a, 15a. This is because the hybrid scheme is not entirely reliant on the population of SCs. However, with increasing number of servers, almost all schemes are impacted through varying degrees.

Query and hybrid-based schemes yield higher service availability, reflecting the direct impact than the SC-based scheme. Since service requests are replied from SCs instead of servers directly, increasing the number of servers would only improve service availability when there are sufficient SCs around to register the presence of services in its neighborhood. This explains the behavior and shortcomings of the SC-based scheme.

With increasing hop count, all schemes are affected and all yield higher service availability. Finally, under the presence of mobility, all three schemes start to perform differently. For the query scheme, mobility has little effect on service availability when the percentage of servers is large, as shown by Figures 5a and 5b. This is because wherever a node moves, it is more likely to reach a near-by server due to the large concentrations of servers. However, when there are few servers, mobility can actually yield greater chances of “meeting” a server at times, depending on the migration pattern.

For the SC-based scheme with mobility, service availability reduces under high percentage of servers and SCs. Migration of SCs and servers away from clients will generally reduce service availability. However, it is observed that migration of SCs when the percentage of servers is low yields better service availability. This is because it increases the chances of SCs “meeting” servers. Such observations are also found in the hybrid-based

scheme. Nonetheless, the outcome is highly dependent on migration patterns of SCs, servers, and clients.

4.2 Discussion on Control Overhead of all 3 Methods

Table 4 Observations on Control Overhead

With increasing percentage SCs

With increasing percentage of servers

With increasing hop count

With presence of mobility

Query case

Not Applicable

Control overhead increases slightly for 1-hop but significantly at 3 hops

Control overhead increases, especially at larger hops

Control overhead increases with presence of mobility

SC-based case

Control overhead generally increases

Control overhead increases and more significantly at larger % SCs and servers

Control overhead increases significantly

Mobility generally resulted in an increase in control overhead, more significantly beyond 10% SCs and 10% servers

Hybrid case

Control overhead only increases significantly at high percentage of servers

Control overhead increases but only significantly at high percentage of SCs.

Control overhead increases significantly

Mobility resulted in significant increase in control overhead, especially at high % of SCs and servers

The previous discussion only focuses on performance of service availability but does not examine the control overhead incurred. As shown in Table 4 above, the query scheme yields greater control overhead with increasing servers, hop count, and under the presence of mobility. This implies that the query scheme, while exhibiting robustness, is not bandwidth efficient.

For the SC-based scheme, control overhead is the result of control information exchanged among the clients, SCs, and servers. We observe that increasing hop count increases control overhead significantly since more control packets have to be propagated. Also, with increasing servers but low percentage of SCs, control overhead does not increase significantly. This is because the SC-based scheme primarily centers the service queries among the clients and SCs. Mobility will result in re-attempts of service query and access, explaining why it will increase control overhead.

Finally, for the hybrid-based scheme, increasing percentage of SCs does not increase control overhead significantly since the other overhead contributing factor is when the hybrid scheme resorts to broadcast query. This explains why only when the percentage of servers is large, increasing SCs will increase control overhead. Similar to the other two schemes, increasing hop count and introducing mobility has the impact of increasing control overhead.

Overall, by comparing all schemes in the static scenario, the query-based scheme yields the least control overhead, this is followed by SC-based scheme, and then hybrid-based scheme, as shown by Figures 6a, 10a, 11a, 12a, 16a, 17a, and 18a. This is still true in the case when mobility is


present, except that SC-based scheme yield similar control overhead with the hybrid scheme at very high percentage of SCs and servers, as shown by Figures 6b, 10b, 11b, 12b, 16b, 17b, and 18b.

5 CONCLUSION

This paper addresses three proposed service location schemes for ad hoc mobile wireless networks. The query scheme relies on client-initiated query and prompt server response. The service-coordinator (SC) scheme, however, acts as a broker where servers register their services on the SC and clients query the SC for locations of desired services. Finally, the hybrid scheme combines both the query and SC-based mechanisms to yield added robustness.

The performance of these schemes is evaluated via simulation, under varying conditions: (a) route hop counts, (b) percentage of servers, (c) percentage of SCs, and (d) the presence of mobility. Simulation results show that the query-based scheme, while simple in concept and implementation, does not yield high service availability (SA) when compared to SC- and hybrid-schemes, especially when the percentage of servers is low. At high percentage of servers, both query and hybrid schemes yield comparable SA performance, better than the SC-based scheme, under the presence of mobility. In terms of control overhead, higher overhead is observed when mobility is present and both SC-based and hybrid schemes yield higher overhead than the query scheme.

In conclusion, the SC-based scheme helps to localize queries to neighboring SCs while the hybrid scheme adds robustness of the query scheme to the SC-based scheme. The choice of a particular scheme will depend on the requirements for ad hoc network service provisioning and the behavior of users (for example mobility profiles) and their devices (i.e., device heterogeneity and capability). Achieving an acceptably high service availability while maintaining reasonable control overhead is the goal for future ad hoc network service providers.

REFERENCES

1. E. Guttman, J. Veizades, M. Days, et. al., “Service Location Protocol, Version 2”, IETF RFC 2608, June 1999.

2. Sun Microsystems, “Jini Architecture Specification, Version 1.1”, October 2000.

3. Microsoft Corporation, “Universal Plug and Play Device Architecture Version 0.92”, May 2000.

4. Salutation Consortium, “Salutation Architecture Specification Version 2.1 (Part 1)”, 1999.

5. Bluetooth Consortium, “Bluetooth Specification Version 1.0B”, December 1999.

6. Munindar P. Singh, Bin Yu, and Mahadevan Venkatraman, “Community-Based Service Location”, Communication of the ACM, Vol. 44, No. 4, pp 49-54, April 2001.

7. Guillermo Guichal, and C.-K. Toh, “Service Location Architectures for Mobile Ad Hoc Wireless Networks”, Proceedings of IEEE Personal Indoor and Mobile Radio Conference (IEEE PIMRC), San Diego, 2001.

service location protocols for mobile wireless ad …...service location protocols for mobile...

Documents