CHAPTER 1 Introduction
1.4 Organization
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5
1.3.3 Data Synchronize issue
This is an independent issue in the thesis. Every node can create new data.
If node can’t connect to the remote database server, it just can share the data by the encounter node. If we would like to share this data with every node, we have to sync this data to the remote server. Thus, every node which can connect to the Internet in the network can query this data. Therefore, if we can’t connect to the Internet, we have to ask the node who can connect to for help to sync the data.
The detail will be described in Section 3.4.
1.4 Organization
The rest of this thesis is organized as follows. Chapter 2 introduces related works in common DTN routing protocol and the Query-based DTN routing protocol. In Chapter 3, we present our searching rule, location-based content search approach. In Chapter 4, we present the simulation results. The final, Chapter 5, we concludes this work and discuss the future work of this thesis.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
6
CHAPTER 2
Related Work
In such a variety of DTN routing researches, we focus on query-based routing schemes. And we divided the routing protocol into two parts, Common DTN routing protocol and Query-based DTN routing protocol, the former, we will introduce some common routing approaches, like Epidemic, Spray and Wait etc., and the latter, we will discuss the approaches which are related to our work. Then, the routing protocol will be discussed below.
2.1 Common DTN routing protocol
We can divide into two kinds of routing protocol: Flooding-based routing and Forwarding-based routing.
2.1.1 Flooding-based routing schemes
Flooding-based means node will have a number of the message copies [15], and it will send this copies to all the encounter nodes. We introduce some approaches below. (1) The basic approach, Epidemic [3], whatever nodes encountered, it will send all the messages it have to encounter nodes. It doesn’t have any selection rule,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
7
just sending until the connection dropped. This approach could get high performance about message delivery success ratio and delivery latency, because a message will be replicated to many nodes, it has high opportunity to send to the destination. But, it will make high overhead, because there are too many redundant messages, and if nodes don’t have enough buffer size, it will overflow quickly. In order to overcome this defect, (2) Spyropoulos et al. proposed Spray and Wait approach [4], they limit the messages replication to L copies, and there are two phase: Spray phase and Wait phase. In Spray phase, when source node encounters another Node B, it will give half copies to Node B to spread these copies. And the other nodes do the same rule, until nodes just have one copy, then they turn to wait phase. In wait phase, all the nodes which have the message copy couldn’t send the copy to another node only if they encounter the destination node. This approach will greatly reduce the transmission overhead, but, it just uses the limitation copies, so the message delivery ratio will be not good as Epidemic. In order to improve this defect, (3) the same authors propose an enhance scheme [5]. They change the wait phase [4] where are routed using direct transmission, they let the waiting nodes have the opportunity to send the message out.
It records every encounter timers, and the timer related a utility value, then use this value to decide whether delivery or not. This rule will keep the low overhead, and enhance the message delivery ratio. There are some approaches [6][12][16] use the history information to predict the delivery probability and decide the transmission. (4) PRoPHET is the representative of this kind of routing approach. Every node has to maintain a delivery predictability table, and every encounter they calculate new prediction value. When nodes encountered, they compare their prediction value, then delivery messages to node have highest prediction value. It gets higher delivery rate than Epidemic, but, because of it spends more time to calculate the prediction value and maintain the predictability table, its delivery latency doesn’t well enough.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
8
2.1.2 Forwarding-based routing schemes
Forwarding-based routing protocols in the network, a message just have one copy [15], and then use different approaches to forward the copy to destination. We introduce some kinds of these schemes, (1) Location-based Routing [17]: it uses GPS to locate nodes position, through different distance algorithm to predict the probability of the message forward to the destination. When nodes encountered, they will compare which one has the highest probability to get to the destination position, then carry it to forward. (2) Gradient Routing: this approach give every node a weighted value, and it represents whether this node suitable for sending message to receiver or not. Different approaches have different way to calculate weighted value. ZebraNet [19] uses history of nodes tracking to get the weighted value, then determine message deliver. In [18], they use social network concept to classify their network node like bubbles, and use the ranking value to decide the messages to.
2.2 Query-based DTN routing protocol
In the past research, many papers [3-6, 15-20] focus on how to send data from source node to destination node, and it doesn’t matter the message response problem.
It is the one way transmission protocol. If a node sends a request to destination node, it wouldn’t know the message being delivered or not. Therefore, one way transmission is not enough, we have to consider how to response the message that we can look up some information. Several works focus on query in DTNs [11-14, 22], in these papers, not only nodes send request to the message owner, but the owners have to send the response message back. The challenge is all nodes are moving and we don’t know where they going, so asker is difficult to get the answer. We introduce four
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
9
works about this kind of request and response pattern research. Then, it describes below:
2.2.1 Locus: A Location-based Data Overlay for Disruption-tolerant
Networks [11]
This research proposed a query mechanism, and use three-tier location area concept (see in Figure 2.a). Let users could look up the relevant information according to its position. Every data object will be given a utility value, and this value will vary with the distance between the creation position of data object and the position of the node which carry this data. Figure 2.b shows the changing of the utility value. If data object near its creation location, the utility of that data object will be high. When nodes encountered, they will change their data objects. Before changing objects, node sorts their data objects by utility value. And the data objects with the higher utility value will be sent first. It is because these data objects were created near here. And this rule would let the data objects be centered on its home area, then, users could query data quickly. But it just use the distance between node and home area, if a node will go out this area, but it is near the home area, encounter node still send data objects to it, then, the transmission is waste. And Locus doesn’t propose an information reply approach, it just uses the multi-copy routing [4][15] to reply data objects.
‧
hemes for D
This resea a copy, Que endliness M ervation wi licate schem
ountered N ue greater th
ives a data ies until it h pose two s ghborhood des encount
ue, the other
a object utili
formance E
DTNs [12]
arch propos ery copy an Metric (FM,
indow) to d me, like the
ode B, Nod han Node A copy of the has only on schemes, W
query spray tered, they
r node gave
ity region Figure 2:
Evaluations
sed a comp d Data reply
it is the av decide the d
e spray phr de A will co A’s, it indicat K copies to ne copy. Thi WSS (W-cop
ying). In W compare th e half query
10
Locus appr
s of Data-ce
plete appro y. The main verage num data delivery
rase of Spr ompare the ates Node B copies to it
b. U roach [11]
entric Infor
ach of data n contribute mber of nod
y. In Data c ve query sp query have ue first, wh t, and then k
Utility Func
rmation Re
a search ste e of the rese
es it encou copy, they p ait routing.
with Node nter more n A follows t g rule. In Q praying) an e w numbe hich one ha keep spread
ction
etrieval
eps, it inclu earch is they unters durin
propose K-c When Nod B, If Node nodes, and N
this rule to Query copy, nd LNS (L
er copies, w ave the hig ding until w
‧
a, this appr erier. This ld get the a ue might be2.2.3 Coo
This pape me nodes ca ess Internet a metadata r wants to q fragments.
ldn’t return
single-copy t compare t or query da s decreased roach [12] j s research a
accurate FM inaccurate.
perative Fi
er introduce an use Wi-F t. Every dat
to descript query a data
Like BitTo n to the origi
F
y scheme, t the FM valu ta first, if n
once. Until just use the ssumes the M value. If . Then these a, it has to orrent [23]
inal data.
Figure 3: Hy
11
g in Hybrid
d DTN routi connects t ivided into e metadata find all the shares rule
ybrid DTN e
L hops to li Node A enc A sends the o
0, it doesn’t T [6] routin ng in a fixd ve in an un
d Deelay To
ing scheme to the Intern
many fragm will be spre metadata o e. But, if o
example [13
imit to que countered N only one qu t spread any ng scheme t d moving pa
npredictable
lerant Netw
e in file sha net, and som ments. And
ead out the of the data t one fragmen
3]
ery spread.
Node B, Nod uery to Nod ymore. In R to reply dat attern, so n e path, the
works [13]
aring. In DT me nodes c
every fragm network. W to downloa nt misses, n
The
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
12
2.2.4 Searching for Content in Mobile DTNs [14]
This research focuses on query how to forward and how many queries are replicated. They propose an estimation rule to avoid the query replicates too much.
When Node A encountered Node B, Node A will estimate the query had been replicated numbers first. The estimate value is calculated by query hop count and network nodes, then, Node A could get a Utility value ( ), if it greater than a threshold T, Node A could send the query to Node B. If not, Node A couldn’t.
Therefore, if this approach can estimate the query replication number accurately, it will greatly decrease transmission overhead and look up data efficiently. But, they just use common multi-copy routing schemes to reply data. And the simulation result, they don’t have compare with other approaches, so we can’t know how this research performed.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
13
CHAPTER 3
Location-Based Content Search Approach
We proposed Location-Based Content Search Approach (LCS) in hybrid DTN to solve the query problem in case of disruptive Internet connection. In our research scenario, all nodes in the network are divided into two types, (1) Online-node and (2) Offline-node. Online-node can connect to the internet to access the remote server, and Offline-node cannot connect to that. All nodes have own database and it includes four tables (see Figure 4), (a) Data Message Table、(b) Encounter Nodes Table、(c) Query Message Table and (d) Reply Message Table. Then we describe below:
(a) Data_table: This table is in charge of the data message storing which is created from node self or received from the other nodes. DataId is the data message unique id, UID is the node id who created this data message, Time is the message creation time or replication time, Data is this message’s content, Sync is whether this message synchronized to the remote server or not. The detail will be described in Section 3.1.
(b) ENs_table: This table is responsible for recording encounter node. UID is the node id of the encounter node, Time is the encounter time, Longitude and Latitude is the encounter location.
(c) Query_table: This table is in charge of the query message storing, if node is
‧
line-node, a ry message yMsg is the ail will be d (d) Reply ated from th que id of thi he reply dat l be describe
Q
In this the d by the o
all query m e unique id e content, is described in y_table: this he node tha is message, ta content, D ed in Sectio
Data_table
Query_table
esis, our go ther nodes
essages are d, Time is t sQueried is
Section 3.2 s table is in at received q Time is thi
e received f this messag whether th 2.
n charge of query mess is message c
the unique
4: Database
the Offline s possible.
from other ge creation his message
f the reply age and had creation tim id of the d
R
e Tables
e-node could In order to
Offline-nod time or re e queried d
message st d match dat me or replica estination n
ENs_table
Reply_table
d get the in o reach this
de. QryId is eplication t data or not.
toring whic ata. RepId is ation time, D node. The d
‧
posed four ery replicat date.
In this the ee parts, Ins
Inside Are ea meters des in differe
er to identif ail see the [ Latitude tw urate to 0.5 P) to calcula
strategies tion strategi
esis, we ass ide Area, B
ea is within s is Border A
ent area use fy the node [26]), this fo wo points to millimeter ate the distan
to solve th ic, (3) Data
sume the ar order Area
Figure 5: T
n a radius o Area and th e different w
position, w formula con o calculate . We use th nce d, if
15
his problem a reply stra
rea range is and Outside
Three parts
of γ meters he other area
ways to sele we refer to th nsidered the
the shortes he node posi
≤ , it indi a is Outside ect message he Vincenty e earth is ell t distance s ition point a icates the no
a replication Data synch
d we divide it shown as
central, ou e Area. This es and deliv y’s formula[
lipsoid, and s, and the d and the Are ode is in the
n strategic, hronization
ed the area s Figure 5.
utside the In s intention i ver message [26](Formu d use Longi deviation ca ea Central P
e Inside Are , (2)
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
16
> ≤ + , it indicates the node is in the Border Area, if > + , it means the node is in the Outside Area.
= ( − ∆ ) (1)[26]
When two nodes encounter, they will change their metadata first (shown as Table 1), it includes node’s position, node’s type (Online-node or Offline node), node past moved path point, node encounter nodes, node collected Data message index list, Query message index list and Reply message index list.
Table 1: Metadata format
Category Value Position Longitude, Latitude
TYPE {0,1}
Path {Pt-1, Pt-2}
Encounter {UID1,UID2,…..,UIDn}
Data list {DataId1_UID1_Time1_Sync1,……,DataIdn_UIDn_Timen_Syncn} Query list {QryId1_Time1_isQueried1,…., QryIdn_Timen_isQueriedn} Reply list {RepId1_Time1, RepId2_Time2,….., RepIdn_Timen}
When two nodes encountered, we call ourself as Node A, encounter node as Node B, the Location-based Content Search approach (LCS) workflow is shown as Figure 6, and the transmission protocol is shown as Figure 7. First, Node A and Node B always detect connection, when encountering, Node A will send a metadata request to Node B, after receiving the request, Node B returned its metadata to Node A, when Node A received the metadata from Node B, the first step is check and filer the identical messages from Node A’s database, then keep the message which Node B doesn’t have, and rebuilt that metadata. The next step is Message Selection, it selects the messages which can be sent to Node B. There are three strategies step we
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17
proposed to select the messages, (1) Data replication strategic, Node A selected the Data messages according to Node A’s position. The detail description is in Section 3.1.
(2) Data Reply strategic, get the Query message from metadata, if Node A’s database have the queried data message, Node A will generate a Reply message. The detail description is in Section 3.2. (3) Query replication strategic, this strategic must be handled after Data Reply strategic complete, and the queries which queried data shouldn’t send out anymore. The detail description is in Section 3.3. After completing three message selection steps, Node A has chosen the messages which will be sent to Node B. The next step, Node A will check the Data messages which doesn’t synchronize to the remote server. If Node A is an Online-node, it will sync this Data messages to the server, if not, Node A doesn’t do anything, and the detail will be described in Section 3.4. Before sending messages to Node B, Node A has to schedule to messages which will be sent first. Because of the contact time of two nodes are short, it couldn’t send all the messages to the other nodes. Therefore, how send the messages to the encounter node efficiency, it is shown in Section 3.5. Then, Node A sends the selection message queue to Node B. As Node B receiving, it will check the Data messages whether it sync to server or not. If Node B is an Online-node, then sync the message to server, if not, skip this action, and then finish this contact.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
Figure 6: LCS workflow
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
Figure 7: LCS transmission protocol
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
20
3.1 Data replication strategic
When two nodes encountered in the map, the processing step can be seen in Figure 8, and the selection rule is shown in Formula 2, (1) If the distance d between Node A and Area Central Point (CP) is less than or equal to radius γ, it indicates Node A is in the Inside Area. When Node A encountered Node B, it will add all the Data messages into the Data dataset D_set to prepare to send to the Node B. The main purpose is let the Inside Area fulfill with related Data messages. If any node wondered to query in this area, it can get the Data messages quickly and has high query success rate. But the encounter time of the two nodes and the transmission are limited, two nodes have to change itself metadata before transport. Then, Node A skips the identical Data messages, and adds the others into D_set; (2) If the distance d between Node A and CP is greater than radius γ and less than γ + , it indicates Node A is in the Border Area. According to Node B’s past path, if ∆ . < 0, it indicates Node B will likely enter to the Inside Area, then, Node A add the Data messages into the D_set; (3) if the distance d between Node A and CP is greater than γ + , it indicates Node A is in the Outside Area, then, Node A shouldn’t do anything.
We use ∆ . to predict whether Node B’s will enter to the Inside Area or not.
In Formula 3, . is the distance between Node B position Pt-1 and CP at time t-1, and . is the distance between Node B position Pt-2 and CP at time t-2.
Then, subtract . from . , if the result is less than 0, it means Node B is likely to enter to the Inside Area, if the result is granter then 0, it means Node B is likely to move away from the Inside Area.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
21
( , _ ) =
, . ≤
, . > . ≤ +
2 ∆ . < 0
, ℎ
(2)
∆ . = ( . − . ) (3)
In order to avoid data overflow, we set all the Data messages a storing time TTS (Time-to-Store) (Formula 4). TTS means the time of any message be created or replicated then store in some node. When TTS expired, the messages will drop anymore. Therefore, it can reduce the node loading.
( ) = , _ _
, ℎ (4)
Otherwise, if Node B is an Online-node, Node A doesn’t need to send Data
Otherwise, if Node B is an Online-node, Node A doesn’t need to send Data