於無線隨意網路中佈建路徑感知之移動節點進行資料重新導向

國

立

交

通

大

學

資訊科學與工程研究所

碩

士

論

文

於無線隨意網路中佈建路徑感知

之移動節點進行資料重新導向

Route-Aware Mobile Relay Deployment for Traffic

Redirection in a Wireless Ad Hoc Network

研 究 生：許修齊

指導教授：曾煜棋 教授

Route-Aware Mobile Relay Deployment for Traffic

Redirection in a Wireless Ad Hoc Network

Student: Hsiu-Chi Hsu

Advisor: Prof. Yu-Chee Tseng

Department of Computer Science

National Chiao-Tung University

於

於

於無

無

無線

線

線隨

隨

隨意

意

意網

網

網路

路

路中

中

中佈

佈

佈建

建

建路

路

路徑

徑

徑感

感

感知

知

知

之

之

之移

移

移動

動

動節

節

節點

點

點進

進

進行

行資

行

資

資料

料

料重

重

重新

新

新導

導

導向

向

向

學生：許修齊 指導教授：曾煜棋教授 國立交通大學資訊科學與工程研究所 摘 摘 摘 要要要 近年來，相當多無線隨意網路相關的研究議題專注在利用可控制移動節點做為網路 中的資料導向裝置以提升網路效能。 本篇論文的研究在混合式的無線隨意網路環境下 提出一個資料重新導向的設計架構； 考慮網路中存在一般的靜止節點，透過加入有限 個數且電量較為充足的可控制移動節點， 幫助原有的資料流量路徑以移動節點作為捷 徑進行適當的資料重新導向。 本篇研究的主要目標在於：透過移動節點進行資料重新 導向的幫助下，減輕網路中所有靜止節點總電量的消耗代價， 我們將此問題定義為移 動節點佈建問題，並設計出一個分散式資料重新導向協定； 在我們的分散式協定中， 移動節點將主動收集網路上的資料流量路徑資訊，計算出最佳的移動路徑， 並與靜止 節點協同合作進行資料重新導向的動作。 最後，我們以QualNet模擬器對於網路效能做 各項指標的評估，模擬結果顯示我們提出的資料重新導向協定能幫助網路環境在 不影 響網路流量的前提下顯著的降低電量消耗。 關鍵字： 控制移動性，無線隨意網路，轉送節點佈建，省電，資料重導。

Route-Aware Mobile Relay Deployment for Traffic

Redirection in a Wireless Ad Hoc Network

Student: Hsiu-Chi Hsu Advisor: Prof. Yu-Chee Tseng

Department of Computer Science National Chiao Tung University

ABSTRACT

Recently, many researches in wireless ad hoc networks focus on using mobility on con-trollable nodes to serve as special roles such as relay nodes to help improve the network per-formance. This paper considers a hybrid wireless ad hoc network and proposes a redirection scheme such that static network nodes can utilize limited number of mobile nodes ,which are resource-rich, as shortcut nodes of existing active flow paths for relaying traffics. Our goal is to mitigate the total energy costs consumed by static nodes with the assistance of mobile nodes. We refer to this problem as the mobile relay deployment problem and develop a novel distributed redirection protocol that mobile nodes actively collect the underlying routing in-formation, optimally define their movements, and cooperate with static nodes in redirecting traffics. Finally, performances with different metrics are evaluated via QualNet simulations, and the simulation results indicate that the proposed protocol results in significant total energy reductions with comparable throughput under variant network environments.

Keywords: Controlled mobility, Wireless Ad hoc Network, Relay deployment, Energy Sav-ing, Flow Redirection.

誌

謝

於兩年研究所期間，受到許多人的關心與幫助使我能順利完成本篇論文，在此表達 我的謝意。 首先，誠摯地感謝我的指導教授曾煜棋老師，除了專業領域知識的獲得之 外，於老師身上感受到 對於工作的熱誠以及觀察問題的深度，引領我在研究的過程中 學習如何面對、解決問題。 並感謝論文口試委員王素華教授、王文良教授以及黃貞芬 教授，於口試時給予寶貴的建議與提醒， 使這篇論文更加完整。 此外，特別感謝實驗室同組的芳璟學姐對於研究上的指導幫助，無論在知識的分享 與討論、 研究的規畫以及論文的寫作上均使我獲益良多。 也由衷的感謝HSCC實驗室 的全體成員，與大家相處的時光豐富了這兩年的研究生活； 以及感謝系足的夥伴們， 我十分珍惜和大家踢球的回憶。 最後，感謝我的父母以及兩位姐姐，在我求學期間給予的支持與關心，使我在學習 的路上無後顧之憂 ，並順利完成學業。感謝所有關心與幫助過我的人。 許修齊 於 國立交通大學資訊科學與工程研究所 中華民國九十九年七月

Contents

摘 摘 摘要要要 I Abstract II 誌 誌 誌謝謝謝 III Contents IV List of Figures VI

List of Tables VII

1 Introduction 1

2 Related Works 5

3 Mobile Relay Deployment Problem 7

4 A Distributed Protocol for the MR-MRD problem 10

4.1 Overhearing Operation . . . 10

4.2 Serving Operation . . . 13

4.2.1 Relocation Phase . . . 13

4.2.2 Service Phase . . . 17

5 Simulation Results 18

6 Conclusion 25

Bibliography 26

List of Figures

1.1 A mobility-aided system for a disaster area. . . 2

1.2 Different relay deployment schemes: (a) Original flow paths. (b) Flow paths redirected by relay nodes in link level. (c) Flow paths redirected by relay nodes in route level. . . 3

3.1 Some scenarios of using mobile nodes to help relay a flow’s traffic. . . 7

4.1 Two simultaneous operations: overhearing and serving. . . 10

4.2 Established routes for the multi-hop flow from S to D. . . 11

4.3 Examples of possible regions. . . 15

5.1 Results with varying number of relay nodes in static networks. (a) Performance Improvement. (b) Throughput. . . 21

5.2 Results with varying network density in static networks. (a) Performance Im-provement. (b) Throughput. . . 22

5.3 Results with varying service length in static networks. (a) Performance Im-provement. (b) Throughput. . . 24

List of Tables

Chapter 1

Introduction

A wireless ad hoc network is a decentralized wireless network that communications occur be-tween any pair of nodes in the absence of any preexisting infrastructure. In such network en-vironments, each node not only operates as a host but also participates in the route discovery process of an ad hoc routing protocol and forwards packets for other nodes in the network that may not communicate directly within the transmission range of each other. Besides, the tasks performed by each node are variant and without fixed schedules, so the traffic pattern and the topology of a ad hoc network may change dynamically.

Recently, controlled mobility on nodes has been proposed as a possible solution that these nodes can act as special roles such as relay nodes to help the network prolong the lifetime. The movement of these relay nodes can be controlled by the underlying protocol to help the network improve the performance with specific objectives. Such mobility-aided system can be applied in several realistic scenarios. Consider a scenario of a wireless ad hoc network constructed in a disaster area. In such area, rescue teams may be dispatched to different small regions for searching survivors and set up communication stations, and some first-aid stations may be established for treatment. These communication systems form a temporarily wireless ad hoc network which is shown in Fig. 1.1. During the period of rescue, multi-hop communications may occur between any source-destination pairs such as one rescue team communicates to the

other team in separate regions or seeks adjacent first-aid stations for support. We can deploy a limited number of mobile robots, which have rich energy or can be charged easily, to serve as guards moving around the disaster area. When the robots detect any active traffic within their communication range during the patrol, they can help relay the signals between the communi-cation pairs. In such mobility-aided system, the total energy consumption of the network nodes can be reduced by the assistance of these relay nodes and therefore the working time of the communication devices are extended.

    
  
        
  
  
   
  
  

Figure 1.1: A mobility-aided system for a disaster area.

In [11], Venkateswaran et al. proposed a relay deployment framework under the environ-ment of an ad hoc network which explores the possibility of using controllable relay nodes to relay the active single-hop flows within its neighborhood, and thus minimizes the energy consumption for transmission in the network. It considers a controllable transmission power scheme for each node in the network since a node consumes much more power in transmitting packets than in the packet reception and idle periods [2]. In their scheme, the sender nodes

uti-
                      
    
       !" #"
    
    

Figure 1.2: Different relay deployment schemes: (a) Original flow paths. (b) Flow paths redi-rected by relay nodes in link level. (c) Flow paths rediredi-rected by relay nodes in route level. lize the relay nodes as intermediate nodes to transmit data if the distances between the senders and the relays are shorter than the original distances of the links, hence the transmission en-ergies can be reduced. However, the design of [11] only considers the active flows in the link level; that is, a multihop flow is decomposed as a sequence of one-hop flows along the multihop path and ignores the routing information on the path to solve the problem.

In this paper, we present a new scheme that exploits the underlying routing information for relaying traffics, and thus works more efficient. We also consider a controllable transmission power scheme to dynamically adjust power between the transmission pairs. The basic idea behind the proposed scheme is to let relay nodes initiatively collect the underlying routing in-formation, detect the active multihop end-to-end flows in the network environment and relocate to the optimal locations such that non-relay nodes can utilize them as shortcuts of original rout-ing paths for communication. Fig. 1.2 shows a clear difference between link level and route level scheme. In Fig. 1.2 (a), There are two original multihop flow paths in the network without

deploying any relay nodes. In the view of link level, every one-hop flow is seen as independent communication pair, so some successive one-hop transmissions in these flow paths are aided by the relay nodes in Fig. 1.2 (b). In contrast, two flow paths are both shortened in Fig. 1.2 (c) since the underlying routing information has been considered and therefore the total transmis-sion energy cost is significantly reduced. To sum up, our goal is to shorten the routing path of the existing multihop flow paths by the assistance of relay nodes rather than finding new routes for the flows, and the extended network lifetime can be achieved since the total transmission energy is minimized by the optimized routes.

The rest of this paper is organized as follows. We discuss the related works in Section 2 and formally define the mobile relay deployment problem and the objective function in Section 3. We present a distributed protocol to the proposed problem in Section 4 and extend the protocol under the consideration of a fully mobile environment. Section 5 evaluates the performance through simulation results, and Section 6 concludes the paper.

Chapter 2

Related Works

Many existing works consider conserving transmission energy as the objective to design the protocols such as power-aware routing with different energy concerns [3][8]. In [3], the authors consider a dynamic power controlled routing scheme. The proposed routing protocol use more intermediate nodes to redirect the original one hop transmission and thus reduce the overall transmission energy consumption. [8] proposed a greedy localized strategy that source node (or intermediate node) selects one of its neighbor to forward packets with minimum total trans-mission power to the destination node. Most of these works joint the concept of redirection by network nodes to the routing algorithms. Variety of issues about power-aware routing protocols in ad hoc network have been surveyed in [4].

Most researches work on utilizing special devices with mobility such as mobile collectors [5][7][14] or mobile relays [12][13] to help lessen the overhead on relaying data along the routing path to the sink in static wireless sensor networks. In [5][7][14], mobile collectors move through the elected points with predefined trajectory to collect data from static sensors in single hop transmissions. [12][13] investigate a heterogeneous sensor network composed of a few mobile relays and static nodes. The authors proposed a joint mobility and routing algorithm that uses mobile nodes to help relay traffics from static nodes and thus improve the network lifetime. However, most of these works are designed for the unique traffic pattern in

wireless sensor network (i.e. all traffics in the network will aggregate to the sink), and may not suitable to a general traffic pattern. Some of these approaches pre-plan the trajectories of mobile nodes; hence, they cannot handle the dynamics in the network environment.

Controlled mobility has been exploited to optimize energy consumption under the environ-ment of wireless ad hoc network as well. In [9], the authors proposed a mobility controlled framework that relocates the network nodes to form an optimal routing path of a existing flow with two objectives: minimize total communication energy consumption and maximize network lifetime. The presented approach considers the energy cost on both communication and node movement for mobility decision making. However, the redirection scheme proposed in this paper is different from [9] since we consider a hybrid network consist of resource-rich mobile nodes to leverage the transmission overhead on simple network nodes. Further, our framework aim to not modify the underlying routing path of existing flows and ensure that the original routing mechanism can still work after mobile nodes stop serving.

Chapter 3

Mobile Relay Deployment Problem

We consider a hybrid ad hoc network with a collection of static nodes and some energy-rich mo-bile nodes. The main function of these momo-bile nodes is to serve as relays to forward those static nodes’ traffics so as to lengthen the network lifetime. We refer to the problem of determining the locations of the mobile nodes the Mobile Relay Deployment (MRD) Problem.


                    
      
   dst. src. m₃ m₂ m₁

Figure 3.1: Some scenarios of using mobile nodes to help relay a flow’s traffic.

Specifically, we are given an ad hoc network with a set NS of static nodes and a set NM

of mobile nodes. The mobility of NM is controllable. Nodes all have the same transmission

distance and are all equipped with GPS receivers, so their locations are always known. Traffics are only generated by static nodes. Let F be a set of multi-hop flows among nodes in NS. Each

flow φi ∈ F is a sequence of nodes in NS between a source-destination pair associated with

a data rate λi. For any two static nodes nj, nk ∈ φi, nj is called nk’s predecessor if nj is the

immediate upstream node of nkin φi, denoted by pre(nk). The main function of NM is to relay

consider any multi-hop flow φj = (n1, n2, . . . , nk). Suppose that a mobile node m ∈ NM is

moved within the transmission ranges of ni and ni+2 ∈ φj. Then we can replace ni+1 ∈ φj

by m and redirect flow φj by a new flow φ0j = (n1, n2, . . . , ni, m, ni+2, . . . , nk). Note that the

location of m is not necessarily the same as the location of ni+1. It is also possible to use more

than one mobile node collaboratively to redirect a flow. Fig. 3.1 shows two examples.

We assume that nodes of NM are much more energy-rich, so we will only focus on the

energy consumption of NS. To model the energy-saving factor, suppose that (ni, ni+1) is a

wireless link of a flow φj ∈ F such that ni is the transmitter and ni+1 is the receiver. The

transmit cost of ni with respect to link (ni, ni+1) of φj is written as Et(λj, ni, ni+1) = λj ·

et(dni,ni+1), where et(dni,ni+1) is the transmit energy function. The energy function et is given

by et(dni,ni+1) = a + b · (dni,ni+1)

α_{, where d}

ni,ni+1 is the distance between niand ni+1, and a, b

and α are environment-related constants (normally α ≥ 2) [8]. On the other hand, the receive cost of ni+1with respect to (ni, ni+1) of φj is written as Er(λj, ni, ni+1) = λj · er, where er is

the receive energy coefficient. Note that since Et is distant-dependent, the location of mobile

nodes would affect the transmit costs of static nodes.

Given NS, NM, and F , our goal is to determine the locations of mobile nodes in NM to

redirect as much traffics of F as possible to save static nodes’ energies. We define the Mobile

Relay Deployment (MRD) Problem as an optimization problem. Let F0 _{be the new set of flows}

after redirection (note that sources and destinations of flows cannot be changed). The original energy cost of F can be written as

E(F ) = X ∀φj∈F X (ni,ni+1) ∈φj (Et(λj, ni, ni+1) + Er(λj, ni, ni+1)) (3.1)

On the contrary, for F0_{, we only include the costs incurred on N} S, so we have E(F0) = X ∀φ0 j∈F0 X (ni,ni+1) ∈φ0 j (b(ni, NS) · Et(λj, ni, ni+1) + b(ni+1, NS) · Er(λj, ni, ni+1)), (3.2)

where b(x, NS) is a binary function which returns 1 if node x ∈ NS, and 0 otherwise (this is

to exclude the cost on NM). The Maximum Reduction MRD (MR-MRD) problem is to find the

Chapter 4

A Distributed Protocol for the MR-MRD

problem

In this section, we propose a distributed protocol for the MR-MRD problem that would work in tandem with the underlying routing protocol adopted by the static nodes. Our protocol has two concurrent operations: overhearing and serving. The overhearing operation is for mobile nodes to collect the routing information in the network. The serving operation, which loops in three states, is to repeatedly relocate mobile nodes to better locations and to help redirect static nodes’ flows. Fig. 4.1 shows the overall operations. More details are given in the subsequent sections.
    
          
 
  
   

   

   

  
 

    
 Ͳ                  

   
 

Figure 4.1: Two simultaneous operations: overhearing and serving.

4.1 Overhearing Operation

In order to find out the opportunity for traffic redirection, each mobile node actively overhears packets sent by its neighboring static nodes. Form the collected information, a Flow Table is maintained, which will help find shortcuts in the serving operation. The table will contain

both active and idle flows passing through. In the following, we assume that static nodes adopt

Ad-hoc On Demand Distance Vector Routing (AODV) [6] as their underlying routing protocol.

In AODV, the path discovery process will issue Route Request (RREQ) to search for routes to intended destinations and Route Reply (RREP) to confirm selected routes. Also, when a route is found to be broken , Route Error (RERR) will be sent along the segmented routes to report the breakage. Each entry in a Flow Table has the format (Ntarget, Nf rom, Chop, Cpkt, Bpkt, τpkt),

where Ntarget is an end node that a flow is connected to, Nf rom is a static node on the flow

neighboring to the mobile nodes, Chop is the hop count from Nf rom to Ntarget, and Cpkt is the

number of packets sent by Nf rom for the flow recently; The last two columns, Bpkt and Tpkt

are used to estimate the data rate for the flow to Ntarget, where Bpkt is the accumulated packet

size and Tpktis the first packet arrival time. The Flow Table is indexed by (Ntarget, Nf rom). For

example, Fig. 4.2 shows a mobile node M with a flow from S to D passing through. Table 4.1 shows that seven entries are maintained for this flow by M. Below, we discuss how a mobile node Mi maintains its Flow Table when overhearing a RREQ/RREP/RERR packet and a data

packet:

S

n

₁

n

2

n

5

n

₄

n

₃

n

₇

n

6

D

M

 





S

 





D

Table 4.1: Flow table maintained by M in the overhearing operation.

Ntarget Nf rom Chop Cpkt Bpkt τpkt

D n1 4 ci bi τi D n2 3 cj bj τj D n4 2 ck bk τk S n5 4 0 0 0 S n4 3 0 0 0 S n2 2 0 0 0 S n1 1 0 0 0

1. RREQ: A static node Si broadcasts a RREQ in the path discovery process of a multihop

flow (Src, Dst) indicates that it’s exactly Src or it’s an intermediate node on the route to

Src, and the hop distance hi from Si to Src is contained in RREQ. When Mi overhears

the RREQ from Si, it inserts an entry (Src, Si, hi, 0, 0, 0) for the reverse route to Src with

Si as the next hop. Duplicated RREQ from Si to Src will be used to update hi.

2. RREP: A static node Si may unicast a RREP to notify the source Src that a route has

been found to the destination Dst or it’s exactly Dst, and contains the hop distance

hi from Si to Dst in RREP. When Mi overhears a RREP from Si, it inserts an entry

(Dst, Si, hi, 0, 0, 0) for the forward route to Dst with Si as the next hop. Duplicated

RREP from Sito Dst will be used to update hi.

3. RERR: A static node Si issues RERR to the previous hop of the route to the destination

Dst for reporting that the broken next hop Sb to Dst is detected in its neighborhood.

When Mi overhears a RERR from Si, it simply invalidates the entries with Dst as Ntarget

and Sbas Nf rom.

4. Data packet: When Mi overhears a data packet from a static node Si for the flow to the

destination Dst , it increases the Cpkt of the entry indexed by (Dst, Si) and accumulates

the corresponding Bpkt. τpkt is set if the data packet is overheard for the first time. For

every time interval ∆T so that the accumulation of Cpkt will be affected by time. Mi

resets Cpkt, Bpktand τpkt once every serving operation.

The management policy of the Flow Table in mobile nodes follows the cache mechanism in AODV and thus the routing information maintained in mobile nodes can be up-to-date. Based on the collected information in Flow Table, each mobile node can utilize the entries with the same Ntarget but different (Nf rom, Chop) to find a redirection opportunity for a stable route to

Ntargetwhose stability is evaluated by Cpktamong these entries, and serve as intermediate node

to redirect the flow traffics toward Ntarget.

4.2 Serving Operation

The serving operation of a mobile node Miis divided into rounds. Each round has three phases:

relocation, service and teardown. In the relocation phase, Mi will collect active flows from its

Flow Table and compute the location where it should move to. In the service phase, Mi will

actively notify the static nodes it should served, start redirecting traffic from these nodes and relocate to the location which is computed in the previous phase. In the teardown phase, Mi

will terminate failed redirections when detecting any link breakage.

4.2.1 Relocation Phase

In the first phase, Mi will detect the active flows by checking the Flow Table maintained by

the overhearing operation and finds out redirection opportunities. A flow is seen to be active if some static nodes on the route which pass through Mi’s neighborhood are enough stable for

transmission. This can be done by checking Cpkt of each entry in the Flow Table. We explain

the relocation phase in detail as follows.

First, Mi skip the entries in the Flow Table if the value of Cpkt < ∆S, where ∆S is a

in Ntarget, Mi checks the entries (Dstk, Ni, hi, ci, bi, τi) to find out redirection opportunity for

flows to Dstk. If there exists two entries (Dstk, Nit, hti, cti, bti, τit) and (Dstk, Nir, hri, cri, bri, τir)

that fulfils the following constraints: (1) Nt

i and Nirare both Mi’s active neighbors, (2) hti > hri

and (ht

i − hri) is maximized, then Mi successfully finds a shortcut for flows to Dst from Nitto

Nr

i and relays the traffics by itself. The first constraint is to ensure that Mi can always service

Nt

i and Nirin the period of the service phase since they are both within Mi’s transmission range,

while the second constraint is to ensure that the existence of the redirections shortens at least one hop from the original routing path and finds the maximum saving in terms of hop distance.

(Nt

i, Nir) is called a redirect pair of Mi for flows to Dstk, and its data rate is estimated by

bt

i and τit. Mi maintains a Service Table which contains a redirect pair, and the data rate and

the original transmission cost of serviced flow for each entry. Since Mi may not be able to

communicate to all the nodes on the path of a flow, it’s challenging to get the original energy cost of the subpath from Nt

i to Nir. Therefore, we use the transmission energy cost for full power

times the shortened hop distance (ht

i− hri) to approximate if (hti− hri) ≥ 2. The construction of

redirect pairs is a greedy choice since we only consider the maximum (ht

i−hri) to find a redirect

pair for each Dstkand simplify the complexity for detecting and searching all subpaths.

After collecting a set of redirect pairs, a local view in Mi’s one-hop neighborhood has

been constructed which is given by (LS, Mi, P ) where LS = {x|x ∈ NS, x is Mi’s one-hop

neighbor}, and P is all the redirect pairs associated with their properties in Mi’s Service Table.

The following operation is to determine the solution to MR-MRD by Mi under (LS, Mi, P ).

We solve MR-MRD by dividing into two subproblems. First, we propose two feasible solu-tions under different considerasolu-tions to select a set of candidate locasolu-tions that mobile relays can relocate to. Then we solve MR-MRD based on the set of candidate locations.

Src

_B

Dst

_B

Dst

_A

Src

_A

B

_t

B

_r

A

_t

A

_r

M

_i

B

_t

B

_r

)}

,

{(

B

t

B

r

V

B

_t

B

_r

A

_t

A

_r

)}

,

(

),

,

{(

A

t

A

r

B

t

B

r

V

Figure 4.3: Examples of possible regions.

divide a specific field centered by Mi into several grid points and let the height and width of the

grid be configurable parameters. In such a way, we can rapidly find a finite set of candidates which contains the locations of all the grid points. The advantage of finding grid points is simplicity and without complicated calculation. However, since it’s an approximate method in the absence of considering network properties, the location for Mi chosen from the candidate

set may not be optimal. In contrast to the grid approximation, the second method make a deeper observation for a local view. For a redirect pair (St, Sr) ∈ P , the possible region to locate

Mi is the intersection area between the communication range of St and Sr. Therefore, each

possible region can be represented by a set of redirect pairs. Fig. 4.3 shows some examples of possible regions with two redirect pairs in Mi’s local view. Once we obtain a possible region

optimal location of Mi in σ that minimizes the transmission energy for redirection. Based on

the energy function Etmentioned in the previous section, the energy consumption for σ when

deploying Mion different locations in σ only affect the transmission energy costed by Sti since

the energy for reception remains constant. The transmission energy cost can be formulated as X

(Si t,Sir)∈Rσ

Et(λi, Sti, Mi) (4.1)

To minimize Eq. (4.1), it can be proved that the optimal location of M should be the closest location in σ to the weighted geometric center Gσ of the locations of all Sti, where the weight

of Si

t is assigned to λi and the exponent α in Et equals two. This optimal location can be

exactly found by geometric relations between Gσ, the area of σ and the locations of Sti and

Si

r. We use the solution for α = 2 as an approximation for scenarios where α > 2. By

enumerating all the possible regions, we can find out all possible ways to redirect flows and the corresponding optimal locations for Mi. It’s the most precise way to list all the candidate

locations. Since it’s more complicated, this solution has higher complexity in calculation than the grid approximation.

After obtaining a set of candidates, we solve MR-MRD based on these locations as follows. Given a set C of candidate locations, we define Reduction(c ∈ C) which returns the sum of the maximum reduction cost, a non-negative value, for each redirect pairs that Mi can serve

at c. For example, if there exists a candidate cx associated with a redirect set {(Stk, Srk)}, then

Reduction(cx) = Σ[max(0, Wtrk − Et(λk, Stk, Mi))] where Wtrk is the original cost of the flow

from Sk

t to Srk. We choose the cmax which has the maximum Reduction(cmax) as the optimal

location for relocating Mi with the corresponding optimal redirect pairs. Mi marks the entries

4.2.2 Service Phase

The service phase is used to start and process the redirection. We describe the operation in the service phase as follows. After mobile node Mi calculates the optimal redirect pairs

{(Si

t, Sri) | Sti, Sri ∈ NS} and the optimal location χop during Computation Slot as mentioned

above, Mi initiates control messages, called Redirect Request (ReREQ), to each Sti and

sched-ules the next move to χop. The static node Sti which receives a ReREQ maintains a soft state

in a Relay Table to keep the redirect information and automatically terminates when either the linkage between Mi and Sti breaks or the timer of the state expires. Sti can simply modify the

source route of the received data packets to Mi if the soft state in the Relay Table exists and

start redirecting. When Mi receives a data packet of a flow (Ssrc, Sdst) from St∗ ∈ Sti, it simply

looks up a marked entry which is indexed by (Ssrc, Sdst) in the Service Table, and then relays

the packet to S∗

r according to the entry.

4.2.3 Teardown phase

In the teardown phase, mobile node Miactively terminates the redirections by initiating control

messages, called Redirect Cancel (ReCAL), to the senders which are being involved in the redi-rection paths. The static node which receives a ReCAL expires the corresponding rediredi-rection soft state in the Relay Table and uses the original route as normal. Notice that, it is possible that links break (e.g. inactive Si

r) during the service phase and the breakages cause packets drop due

to failed redirections. Therefore, mobile nodes may send ReCal to the senders which are being involved in failed redirection paths before the teardown phase in such special situations.

Chapter 5

Simulation Results

In this section, we evaluate the performance of our proposed redirection schemes. We imple-mented the relay redirection framework in QualNet simulator [10]. We place network nodes by uniform random deployment in a 750m × 750m 2D terrain. Each simulation run is simu-lated for 1 hour duration and each of the results presented in this section is averaged over thirty simulation runs.

We simulate the ad hoc network under IEEE 802.11b environment with a channel data rate of 2 Mbps and use 802.11 MAC protocol for wireless transmission through a free space path loss model. The maximum transmission power of each node was set to 6.633 dBm which corresponds to a transmission range of approximately 250 meters under a packet reception threshold of -91 dBm. We set the power consumption model in a similar way represented in [1] to simulate a realistic implementation of the network interface. There are three categories of end-to-end traffic flows in the network:

1. Four high-rate 120kbps UDP request traffic with 1 KB packet size. 2. Four medium-rate 40kbps UDP request traffic with 1 KB packet size. 3. Four low-rate 0.8kbps UDP request traffic with 100 byte packet size.

For each request from the source node, the destination node responses one reply packet with 512 byte packet size. All traffic flows are random source-destination pairs with random start

times and durations. Besides, through the simulation, all static nodes in the network broadcast a Beacon message once every Beacon Interval. The beacon is used to exchange location in-formation and preserve connectivity among neighbor nodes. These inin-formation can help static nodes to estimate the distances between neighbors at a specific time and to appropriately adjust their transmit power for communication. We let Beacon Interval be equal to the length of the service phase.

We compare our proposed protocol (denoted by ’MR-MRD’) against the deployment frame-work with minimum total energy (denoted by ’Min-Total’) proposed in [11] and use AODV as the underlying routing protocol in the simulation. The energy reduction in both MR-MRD and Min-Total can be divided into two parts. The first part is from the controllable power scheme which minimizes the total energy consumption by reduce the transmission power according to the distance between the communication pairs; the second part is due to the traffic redirection scheme by controlled mobile nodes. To quantify the part of total reduction by power control, we apply the controllable power scheme to AODV (denoted by ’Aodv-Pc’) and without the assistance of any relay nodes as a comparison protocol. In our simulation results, we use the network environment, which is only composed of non-relay nodes and functions AODV (de-noted by ’Aodv-Pure’), as a basis and design two metrics to evaluate the performance among different protocols as follows.

1. The total transmission energy saving: We measure the energy saving in the network as follows. Let E be the total transmission energy in Aodv-Pure and EP denote the total

energy consumed in transmission when the protocol P is applied. Then the energy saving in transmission is computed as (E − EP)/E.

2. Normalized throughput: We measure the throughput of the traffic flows in a protocol and normalize the value by the throughput in Aodv-Pure.

In the first experiment, we observe the performance when varying number of mobile relay nodes. We fix the number of static nodes and the length of the service phase to 25 nodes and 15 seconds respectively. Fig. 5.1 (a) shows that power control scheme (i.e. Aodv-Pc) reduce about 24 percents of total transmission energy consumption. Our redirection scheme, which can improve the saving to about 28 to 40 percents, is much better than Min-Total. This result shows that redirecting in route level has more opportunities than in link level, and thus has better performance gain. However, the increasing number of additional relay nodes results in the increasing channel contention when a node start to transmit packets and decreases the channel utilization. Fig. 5.1 (b) shows that the network throughput decreases due to the assistance of additional relay nodes rather than power control in transmissions. Besides, when a route entry in the route table is used to transmit a data packet, its lifetime will be extended. The route-level redirection of flows in MR-MRD let some intermediate nodes be omitted in the new path and has less chance to update their entries. Thus, these entries expire quickly during the redirection. Note that route entries will always be updated in link-level redirection since data packets are only forward a single hop. This phenomenon results in more data packets dropped due to expired routes in MR-MRD than in Min-Total after relay nodes leave, and explains the performance gap between them.

 
 
 
    
                       ! " # $ %& '  " ( ) * +, -Ͳ. / 01 Ͳ01 2 03 4 Ͳ5 +67 8 (a)




                            ! " # $ % & ' ( )% * +, ! ' -% . / 01 2 Ͳ3 4 56 Ͳ567 589 Ͳ: 0;< = (b)

Figure 5.1: Results with varying number of relay nodes in static networks. (a) Performance Improvement. (b) Throughput.

In the next experiment, we deploy 4 relay nodes and keep the length of the service phase in 15 seconds. We vary the network density by deploying different number of static nodes and compare the performance. In a sparse network, longer distance between nodes and their neighbors result in more energy consumed on some nodes in data transmissions, and thus the flow traffics relaying by additional relay nodes can significantly leverage the overhead of these static nodes, as can be seen in Fig. 5.2 (a). The networks with variety of density do not affect

the occurrences of redirection since the hop distances of the routing paths selected by AODV remain stable through these networks, and result in a stable trend of throughput which is shown in Fig. 5.2 (b).
 
 
 
  
 
                        ! "#"$ %   & ' ( )* + Ͳ, -./ Ͳ./ 0 .12 Ͳ3 )45 6 (a)




                           ! " #$ % & ' () *)+ ,  & -$ . / 01 2 Ͳ3 4 56 Ͳ56 7 589 Ͳ: 0;< = (b)

Figure 5.2: Results with varying network density in static networks. (a) Performance Improve-ment. (b) Throughput.

To observe the effect of different periods of service length, we deploy 25 static nodes and 4 mobile relay nodes, and vary the length of the service phase. Fig. 5.3 (a) shows that the traffic pattern of the network environment may be suitable for a particular setting of the service phase. Appropriate setting of the service phase can lead to a better traffic redirection; that is, mobile nodes will not stay in the neighborhood of flows with short durations, and wait for service even if the flows become inactive. Fewer redirections occur mitigate the control message overheads in the network, increase the channel utilization and thus increase the throughput of the network. Specifically, static nodes in Min-Total need to initiatively detect active flows and inform relay nodes about these information; therefore, Min-Total has more benefits from reduced control message overheads than MR-MRD in terms of throughput. Fig. 5.3 (b) shows the results for throughput.

 
 
 
 
 
    
                    !   " # $ % &  ! ' ( )* + Ͳ, -./ Ͳ./0 .12 Ͳ3 )45 6 (a)




                                ! " # $ % !  & ' ( ) * ! % + , -. / Ͳ01 23 Ͳ234 256 Ͳ7 -8 9 : (b)

Figure 5.3: Results with varying service length in static networks. (a) Performance Improve-ment. (b) Throughput.

Chapter 6

Conclusion

Energy-conserving is a critical issue in wireless ad hoc networks. We consider controlled mo-bility as a solution that some controllable resource-rich nodes can act as relay nodes to help network lessen the total energy costs. In this paper, we defined the mobile relay deployment problem that aims to minimize the energy consumption at static nodes with the assistance of relay nodes, and proposed a novel distributed protocol that utilizes underlying ad hoc protocol information to optimally shorten the routing paths of existing flows. The simulation results indi-cate that our protocol results in significant total energy reductions with comparable throughput under different network environments.

Bibliography

[1] J. Ebert, B. Burns, and A. Wolisz. A tracebased approach for determining the energy consumption of a wlan network interface. In Proc. European Wireless, pages 230 – 236, 2002.

[2] L. Feeney and M. Nilsson. Investigating the energy consumption of a wireless network interface in an ad hoc networking environment. In INFOCOM 2001. Twentieth Annual Joint Conference of the

IEEE Computer and Communications Societies. Proceedings. IEEE, volume 3, pages 1548 –1557

vol.3, 2001.

[3] J. Gomez, A. T. Campbell, M. Naghshineh, and C. Bisdikian. Paro: supporting dynamic power controlled routing in wireless ad hoc networks. Wirel. Netw., 9(5):443–460, 2003.

[4] J. Li, D. Cordes, and J. Zhang. Power-aware routing protocols in ad hoc wireless networks. Wireless

Communications, IEEE, 12(6):69 – 81, dec. 2005.

[5] M. Ma and Y. Yang. Data gathering in wireless sensor networks with mobile collectors. In Parallel

and Distributed Processing, 2008. IPDPS 2008. IEEE International Symposium on, pages 1 –9,

14-18 2008.

[6] C. Perkins and E. Royer. Ad-hoc on-demand distance vector routing. In Mobile Computing Systems

and Applications, 1999. Proceedings. WMCSA ’99. Second IEEE Workshop on, pages 90 –100,

25-26 1999.

[7] J. Rao, T. Wu, and S. Biswas. Network-assisted sink navigation protocols for data harvesting in sensor networks. In Wireless Communications and Networking Conference, 2008. WCNC 2008.

IEEE, pages 2887 –2892, march 2008.

[8] I. Stojmenovic and X. Lin. Power-aware localized routing in wireless networks. IEEE Trans.

Parallel Distrib. Syst., 12(11):1122–1133, 2001.

[9] C. Tang and P. McKinley. Energy optimization under informed mobility. Parallel and Distributed

Systems, IEEE Transactions on, 17(9):947 –962, sept. 2006.

[10] S. N. Technologies. Qualnet simulator. http://www.scalable-networks.com.

[11] A. Venkateswaran, V. Sarangan, T. La Porta, and R. Acharya. A mobility-prediction-based relay deployment framework for conserving power in manets. Mobile Computing, IEEE Transactions

on, 8(6):750 –765, june 2009.

[12] W. Wang, V. Srinivasan, and K.-C. Chua. Using mobile relays to prolong the lifetime of wireless sensor networks. In MobiCom ’05: Proceedings of the 11th annual international conference on

Mobile computing and networking, pages 270–283, New York, NY, USA, 2005. ACM.

[13] W. Wang, V. Srinivasan, and K.-C. Chua. Extending the lifetime of wireless sensor networks through mobile relays. Networking, IEEE/ACM Transactions on, 16(5):1108 –1120, oct. 2008. [14] G. Xing, T. Wang, W. Jia, and M. Li. Rendezvous design algorithms for wireless sensor networks

with a mobile base station. In MobiHoc ’08: Proceedings of the 9th ACM international symposium

Curriculum Vita

Hsiu-Chi Hsu

Contact Information

Department of Computer Science National Chiao Tung University

1001 Ta Hsueh Road, Hsinchu, Taiwan 300 Email: [email protected]

Education

M.S.: Computer Science, National Chiao Tung University (2008.9 ∼ 2010.6) B.S.: Computer Science, National Chiao Tung University (2004.9 ∼ 2008.6)

Publication Lists

1. F.-J. Wu, H.-C. Hsu, Y.-C. Tseng, and C.-F. Huang, “Non-location-based Mobile Sensor Relocation in a Hybrid Static-Mobile Wireless Sensor Network”, In Proceedings of the 2009