Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
1
19.03.2009 © SUPSI-DTI 1
Dipartimento
Tecnologie
Innovative
Routing in Opportunistic Networks
Silvia [email protected] in collaboration with:
Marco Conti, Jon Crowcroft, Pan Hui, Hoang Anh Nguyen, Andrea Passarella
and the whole Haggle project
thanks to Cecilia Mascolo and Martin May
19.03.2009 © SUPSI-DTI 2
Dipartimento
Tecnologie
Innovative
Outline� Internet: where are we?
� ubiquitous computing/communication
� clean-slate approach
�New Networking Technologies� Haggle
� DTN
� Opportunistic Networks
� Routing in OppNets� context based routing
� C-BR classification
� SANETs
2
19.03.2009 © SUPSI-DTI 3
Dipartimento
Tecnologie
Innovative
Ubiquitous Computing: Business Users
�users can work without interrupting their work even when they move
�users must interrupt their work when they move
19.03.2009 © SUPSI-DTI 4
Dipartimento
Tecnologie
Innovative
Ubiquitous Computing: Other Users�visiting people
�young people
3
19.03.2009 © SUPSI-DTI 5
Dipartimento
Tecnologie
Innovative
First dream
M. Conti and S. Giordano. Multihop Ad Hoc Networking: The Theory.IEEE Communications Magazine, 2007
Example: realize the ubiquitous computing with MANET technology.
19.03.2009 © SUPSI-DTI 6
Dipartimento
Tecnologie
Innovative
Missing a pragmatic approach!
Example: realize the ubiquitous computing with MANET technology.
M. Conti and S. Giordano. Multihop Ad Hoc Networking: The Reality.IEEE Communications Magazine, 2007
4
19.03.2009 © SUPSI-DTI 7
Dipartimento
Tecnologie
Innovative
But pervasiveness is a reality
ChatpenDigital Information recognized and transmitted by penpen
CeivaDigital Picture Frame
Plug &Play640 x 480 VGAJPEG Image FormatStorage: 10 photos/250 on-line
Plug &Play640 x 480 VGAJPEG Image FormatStorage: 10 photos/250 on-line
19.03.2009 © SUPSI-DTI 8
Dipartimento
Tecnologie
Innovative
Clean-slate approach�Heterogeneous networks technologies
�Heterogeneous applications
� New architecture and communications paradigms:� DTNs
� ANA
� Haggle: �opportunistic networks,
� social networks
5
19.03.2009 © SUPSI-DTI 9
Dipartimento
Tecnologie
Innovative
The Internet Hourglass
IP
Applicationlayer
Linklayer
Ethernet, WIFI (802.11), ATM,SONET/SDH, FrameRelay,modem, ADSL, Cable, Bluetooth…
Disruptive approaches need
a disruptivearchitecture
Changing/updatingthe Internet coreis difficult orimpossible !(e.g. Multicast,Mobile IP, QoS, …)
Everythingon IP
IP on
Everything
Homogeneous networking abstraction
Voice, Video, P2P, Email, youtube, ….Protocols – TCP, UDP, SCTP, ICMP,…
IPv6
19.03.2009 © SUPSI-DTI 10
Dipartimento
Tecnologie
Innovative
Approach 1: ANA
�We have to widen the waist and host more paradigms
�Federation instead of homogeneous abstraction
�May the best concept wins
Applicationlayer
Linklayer
We need the framework for these multiple networks
6
19.03.2009 © SUPSI-DTI 11
Dipartimento
Tecnologie
Innovative
ANA ≠ "one-size-fits-all"� A network compartment implements the operational rules and administrative policies for a given communication context
� Publish Subscribe Compartment� PodNet compartment
ANA framework
Application
Linklayer
19.03.2009 © SUPSI-DTI 12
Dipartimento
Tecnologie
Innovative
Approach 2: DTN
�Support interoperabilityacross ‘radically heterogeneous’ networks
�Tolerate delay and disruption� Acceptable performance in high
loss/delay/error/disconnected environments
� Decent performance for low loss/delay/errors
Applicationlayer
Linklayer
Still IP based
7
19.03.2009 © SUPSI-DTI 13
Dipartimento
Tecnologie
Innovative
Delay-Tolerant Networking Architecture*
�Components� Flexible naming scheme
� Message abstraction and API
� Extensible Store-and-Forward Overlay Routing
� Per-(overlay)-hop reliability and authentication
*from DTN group slides
Still IP based
19.03.2009 © SUPSI-DTI 14
Dipartimento
Tecnologie
Innovative
Village
Internet
City
Example Routing Problem *2
3 1
bike
time (days)
Connectivity: Village 1 – City
bandwidth
bikesatellitephone
*from DTN group slides
8
19.03.2009 © SUPSI-DTI 15
Dipartimento
Tecnologie
Innovative
DTNs: What ?�Network with intermittent connectivity
�Data are disseminated exploiting host mobility
� long and variable delays
� high error rates
� heterogeneity
19.03.2009 © SUPSI-DTI 16
Dipartimento
Tecnologie
Innovative
DTNs: Why ?�To cover regions with no infrastructure
�Human social networks
�Intelligent highways
�Emergency rescue
�Sensor networks
�Wildlife monitoring
9
19.03.2009 © SUPSI-DTI 17
Dipartimento
Tecnologie
Innovative
Approach 3: Haggle�new autonomic networking architecture
� enables communication for intermittent network connectivity,
� exploits autonomic opportunistic communications
(i.e., when no end-to-end communication infrastructures).
�support interoperabilityacross ‘radically heterogeneous’ networks
Applicationlayer
Linklayer
No more layers!
Physical Link
ApplicationData
Haggle
19.03.2009 © SUPSI-DTI 18
Dipartimento
Tecnologie
Innovative
Haggle (2006-2010)
� Started January 2006 - EU FP6 – FET� developed 3 prototypes (Infant, Child & Young Haggle)
� code on sourceforge� http://www.haggleproject.org
Give it to me, I have 1G bytes phone flash.
I have 100M bytes of data, who can carry for me?
We can also carry for you!
Thank you but you are in the opposite direction!
Don’t give to me! I am running out of storage.
Reach an access point.
Internet
Finally, it arrives…
100MB
10
19.03.2009 © SUPSI-DTI 19
Dipartimento
Tecnologie
Innovative
� a revolutionary paradigm for communicating without end-to-end connection
� relax the end-to-end paradigm
User
controlHIS
ClassifierForward
802.11 Bluetooth EthernetCradle
Incentive
Authentication
Resource control
GPRS Connectivitymanagement
Incentive
Authentication
Resource controlHagglecontrol
19.03.2009 © SUPSI-DTI 20
Dipartimento
Tecnologie
Innovative
� a context based routing
�fwding manager with more strategies
Search La Bonheme.mp3 for me
Search La Bonheme.mp3 for me
There is one in my pocket…
Sorry, I don’t know any singer called Bonheme!
Uh! I love La Bonheme! (My wife should have it…)
Search La Bonheme.mp3 for me
11
19.03.2009 © SUPSI-DTI 21
Dipartimento
Tecnologie
Innovative
Opportunistic NetworksOpportunistic networking: we are surrounded by communicating devices and most of applications are not real time constrained
�relax the end-to-end paradigm�communication is “on-the-fly”� heterogeneous devices
� mobile phones�laptops, palm�sensors, etc...
� exploit mobility� real best-effort
19.03.2009 © SUPSI-DTI 22
Dipartimento
Tecnologie
Innovative
Opportunistic Networks – The Missing Link?
In oppnets, diverse systems—not deployed originally as oppnet nodes—join an oppnet dynamically in order to perform certain tasks they have been invited (or ordered) to participate in
�Not always connected, “internet connectivity islands”�Huge amount of untapped resources in portable devices
� Local wireless bandwidth� Storages� CPUs
�A packet can reach destination using network connectivity or user mobility
12
19.03.2009 © SUPSI-DTI 23
Dipartimento
Tecnologie
Innovative
CONTEXT
� Oppnets are envisioned to provide, among others:� Bridges between disjoint communication media
� Additional platforms for offloading tasks
� Additional sensing modalities by integrating existing independent sensory systems
� Oppnets can exploit context information
Search La Bonheme.mp3 for me
Search La Bonheme.mp3 for me
Search La Bonheme.mp3 for me
There is one in my pocket…
Objectives
19.03.2009 © SUPSI-DTI 24
Dipartimento
Tecnologie
Innovative
OppNets routing: context information
� the context in which the users communicate. � Examples:
� users' working address and institution, personal information, habits
13
19.03.2009 © SUPSI-DTI 25
Dipartimento
Tecnologie
Innovative
OppNets routing: context information
� Thus:� given context information about the destination
� suitable forwarders can be found based on:� the probability of meeting with other users
� the probability of visiting particular places
I’ve a message for Jack
Jack is at school school bus is just coming
I’ll wait the school bus and give the message to someone in it, who is going to school, who is likely to meet Jack in few minutes
19.03.2009 © SUPSI-DTI 26
Dipartimento
Tecnologie
Innovative
Context aware routing classification
� context-oblivious:� basically exploit some form of flooding
� partially context-aware (aka mobility-based):� exploit some particular piece of context information to optimise the forwarding task.
� fully context-aware (aka social context-aware):� not only exploit context information to optimiserouting, but also provide general mechanisms to handle and use context information.
14
19.03.2009 © SUPSI-DTI 27
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� exploit some form of flooding.
� a message should be disseminated as widely as possible.
� might be the only solution when no context information is available.
� generate a high overhead
� may suffer high contention and potentially lead to network congestion
19.03.2009 © SUPSI-DTI 28
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� To limit this overhead, the common technique is to control flooding � by limiting the number of copies allowed to exist in the network
� by limiting the maximum number of hops a message can travel.
� In the latter case, when no relaying is further allowed, a node can only send directly to the destination when (in case) it is met.
15
19.03.2009 © SUPSI-DTI 29
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� Epidemic Routing � A. Vahdat and D. Becker. Epidemic Routing for Partially Connected Ad Hoc Networks. Technical Report CS-2000-06, CS. Dept. Duke Univ., 2000.
� Network Coding � J. Widmer and J. Le Boudec. Network coding for efficient communication in extreme networks. Proc. of the ACM SIGCOMM WDTN , 2005.
� Spray&Wait� T. Spyropoulos, K. Psounis, and C. Raghavendra. Efficient routing in intermittently connected mobile networks: The multiple-copy case, ACM/IEEE Transactions on Networking, 16, 2007.
19.03.2009 © SUPSI-DTI 30
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� Epidemic Routing � When two nodes meet exchange summary vectors with compact unambiguous representation of the messages currently stored in their local buffers.
� Then, each node requests from the other the messages it is currently missing.
� hop count limits dissemination (hc = 1 -> only to dest.).
A BV[msg]a,b,d
16
19.03.2009 © SUPSI-DTI 31
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� Network Coding � an original approach to limit message flooding.
� In general, network coding-based routing outperforms flooding, as it is able to deliver the same amount of information with fewer messages injected into the network.
A B Ca acc a c a c+ +
19.03.2009 © SUPSI-DTI 32
Dipartimento
Tecnologie
Innovative
Context-oblivious Protocols
� Spray&Wait� two phases: the spray phase and the wait phase.
� spray phase: L copies of the same message are spread over the network
� wait phase: each relay node stores its copy and eventually delivers it to the destination when (in case) it comes within reach.
� L can is chosen based on a target average delay.� spray can be performed in multiple ways (e.g. binary)
17
19.03.2009 © SUPSI-DTI 33
Dipartimento
Tecnologie
Innovative
Partially Context-aware
� exploit some particular piece of context information to optimise the forwarding task.� e.g. mobility of nodes
� Main difference with fully context-aware:� generally is customized for a specific type of context information
� does not provide a full-fledged set of algorithms to gather and manage any type of context information
19.03.2009 © SUPSI-DTI 34
Dipartimento
Tecnologie
Innovative
Partially Context-aware� Probabilistic ROuting Protocol using History of Encounters and Transitivity (PROPHET)� A. Lindgren, A. Doria, and O. Schelen. Probabilistic routing in intermittently connected networks. ACM Mobile Computing and Communications Review, 2003.
� MV and MaxProp� B. Burns, O. Brock, and B. Levine. MV routing and capacity building in disruption tolerant networks. Proc. of the 24th INFOCOM,2005.
� J. Burgess, B. Gallagher, D. Jensen, and B. Levine. MaxProp: Routing for Vehicle-Based Disruption-Tolerant Networks. Proc. of the 25th INFOCOM, 2006.
18
19.03.2009 © SUPSI-DTI 35
Dipartimento
Tecnologie
Innovative
Partially Context-aware� Last Encounter Routing and Spray&Focus
� M. Grossglauser and M. Vetterli. Locating nodes with EASE: last encounter routing in ad hoc networks through mobility diffusion. Proc. of the 22nd INFOCOM, 2003.
� T. Spyropoulos and K. Psounis. Spray and focus: Efficient mobility-assisted routing for heterogeneous and correlated mobility. Proc. of IEEE Percom International Workshop on Intermittently Connected Mobile Ad hoc Networks, March 2007.
� MobySpace Routing� T. C. V. Leguay, J.; Friedman. Evaluating mobility pattern space routing for
dtns. Proc. of the 25th INFOCOM, 2006.
� Bubble Rap� P. Hui and J. Crowcroft. Bubble rap: Forwarding in small world dtns in
every decreasing circles. Technical report, Technical Report UCAM-CL-TR684, Univ. of Cambridge, 2007.
19.03.2009 © SUPSI-DTI 36
Dipartimento
Tecnologie
Innovative
Partially Context-aware� PROPHET
� evolution of Epidemic: during a contact, nodes also exchange their delivery predictability (DP) to destinations of messages they store in their buffers
� messages are requested only if DP is higher than that of the node currently storing the message.
� DP is the probability for a node to encounter a certain destination� increases when the node meets the destination
� decreases (according to an ageing function) between meetings
� context information used by PROPHET is the frequency of meetings between nodes
19
19.03.2009 © SUPSI-DTI 37
Dipartimento
Tecnologie
Innovative
Partially Context-aware� MV and MaxProp
� in addition, also exploit information about the frequency of visits to specific physical places.
� Last Encounter Routing and Spray&Focus� exploit the fact that the decreasing gradient of the time lag identifies a suitable path towards the destination.
� MobySpace Routing� the mobility pattern of nodes is the context information used for routing
19.03.2009 © SUPSI-DTI 38
Dipartimento
Tecnologie
Innovative
Bubble-Rap� automatically inferring the parameters of the underlying
social structure.
� exploiting the structure properties to select paths.
� dynamically identifies users’ communities, and ranks the nodes ”sociability” (measured as the number of links) inside each community.
20
19.03.2009 © SUPSI-DTI 39
Dipartimento
Tecnologie
Innovative
Fully Context-aware
� provide general mechanisms to handle and use context information
� more general than PCA:� works with any context information
� can be customized for the specific environment it must work
19.03.2009 © SUPSI-DTI 40
Dipartimento
Tecnologie
Innovative
Fully Context-aware� Context-Aware Routing (CAR)
� M. Musolesi, S. Hailes, and C. Mascolo. Adaptive routing for intermittently connected mobile ad hoc networks. Proc. of WoWMoM 2005.
� History Based Opportunistic Routing (HiBOp)� C. Boldrini, M. Conti, I. Jacopini, A. Passarella, and P. IIT-CNR. HiBOp: a
History Based Routing Protocol for Opportunistic Networks. Proc. of WoWMoM 2007
� Probabilistic Routing Protocol for Intermittently Connected Mobile Ad hoc Network (PROPICMAN/Spatio-Tempo)� H. Anh Nguyen, S. Giordano, A. Puiatti. Probabilistic Routing Protocol for
Intermittently Connected Mobile Ad Hoc Networks (PROPICMAN). Proceedings of IEEE WoWMoM/AOC 2007.
� H. Anh Nguyen, S. Giordano, Spatiotemporal Routing Algorithm in Opportunistic Networks. Proceedings of IEEE WoWMoM/AOC 2008
21
19.03.2009 © SUPSI-DTI 41
Dipartimento
Tecnologie
Innovative
Context-Aware Routing (CAR)� Nodes are divided in groups
� Nodes inside the same group havesocial links between each other
� Are connected by an underlying MANET routing protocol
� Nodes of different groups can have social links as well
� Nodes reconfigure
19.03.2009 © SUPSI-DTI 42
Dipartimento
Tecnologie
Innovative
CAR� A node selects the fwder node/group
� Probabilistically
� Based on the social attraction among nodes
� multi-attribute utility-based framework is defined to combine context information for computing the delivery probabilities
� CAR compute delivery probabilities proactively, and disseminate them in their ad hoc cloud.
� context information is exploited to evaluate probabilities just for known destinations
22
19.03.2009 © SUPSI-DTI 43
Dipartimento
Tecnologie
Innovative
CAR: conjunction beteewn PCA & FCA � The framework is general enough to accommodate different types of context information.
� Context information in CAR consists of the logical connectivity of nodes, such as the rate of connectivity change, the delivery probability of the neighbor nodes to the destination, and device information, such as residual battery life.
� However, the social context information is not taken into consideration as in social context-based category
� CAR can be seen as a conjunction between mobility-based protocols and social context-based protocols.
19.03.2009 © SUPSI-DTI 44
Dipartimento
Tecnologie
Innovative
44
HiBOp� Mimic the establishment of social relationships among people
� Users describe themselves� Users decide what information about themselves and their habits they want to make publicly available
� users store this information into their own devices
� HiBOp is not bound to any predefined set of information
Personal InformationName DonaldSurname DuckEmail [email protected] 3403434398WorkStreet Via Moruzzi, 1City PisaInstitute IIT-CNR
attributes values
Identity Table
23
19.03.2009 © SUPSI-DTI 45
Dipartimento
Tecnologie
Innovative
45
HiBOp in a nutshell� Sender does several copies (k)of the message
� To achieve atarget delivery probability
� Messages are discarded after being handed over to the next hop� i.e., message spread is controlled by the target delivery probability
k = min K | (1− psucc( i ) )
i =0
K
∏ ≤ plmax
19.03.2009 © SUPSI-DTI 46
Dipartimento
Tecnologie
Innovative
Propicman: Probabilistic Routing Protocol for Intermittently Connected Mobile Ad hoc Network
�Exploit the context information of nodes to select the best next hop candidate.
�Each node has a common node profile withevidence/value pairs.
�Evidences have different weights.
�Compute the deliverity probability to thedestination.
�Zero-conf
24
19.03.2009 © SUPSI-DTI 47
Dipartimento
Tecnologie
Innovative
Propicman social approach� Common set of evidences for every node - T
� Node profile:
W2E2
……
WnEn
W1E1
Weights(W)Evidence names(E)
H(En,Vn)VnEn
………
H(E2,V2)V2E2
H(E1,V1)V1E1
Hashed valuesValuesEvidence names
5City
7Profession
……
2Nationality
10Work place
Weights(W)Evidence names(E)
An example:
14a12de02cd9c0c5fd94…
LuganoCity
536a195c27b76cc2b175…
SwissNationality
………
c056817c8f7177163c83…
Profession
26a48976c47c0d2b2383…
SUPSIWork place
Hashed valuesValuesEvidence names
19.03.2009 © SUPSI-DTI 48
Dipartimento
Tecnologie
Innovative
Sender
A
A1
A2
A3
hM
……
An
hMhMhMhM
PA1
PA2
PA3
PAn
PA*max{PAi}
Propicman
25
19.03.2009 © SUPSI-DTI 49
Dipartimento
Tecnologie
Innovative
Propicman� Neighbors do local matching between received hM with its node profile.
� Then neighbor node (A) can compute its PA:
0≤ PA = ≤1
� : sum of the weights of all the evidences that S knows about D.∑
∑Mi
i
W
Wmatched )(
∑ iMW
H(ED1,VD1) H(ED2,VD2) H(ED3,VD3) H(ED4,VD4) H(EDn,VDn) MACS SN
Message header hM
…
Hashed values in node profile
H(E1,V1) H(E2,V2) H(E3,V3) H(E4,V4) H(En,Vn)…
19.03.2009 © SUPSI-DTI 50
Dipartimento
Tecnologie
Innovative
Morning
26
19.03.2009 © SUPSI-DTI 51
Dipartimento
Tecnologie
Innovative
Afternoon
19.03.2009 © SUPSI-DTI 52
Dipartimento
Tecnologie
Innovative
Evening
27
19.03.2009 © SUPSI-DTI 53
Dipartimento
Tecnologie
Innovative
One cycle
a period
7am-8am 8am-9am 9am-12am 12am-1pm 1pm-5pm 5pm-8pm 8pm-7am
Similarity!
19.03.2009 © SUPSI-DTI 54
Dipartimento
Tecnologie
Innovative
SpatioTempo – Design
�Each period can have different contact opportunities. In some periods source nodeencounters few devices, some encounterbunch of devices with high DP.
: period ith of the cycle c-1
: set of encountered nodes in period ith of the cycle c-11−ciS
1−ciT
11−
+c
iT1−ciT
One cycle
ciT 1−
1−ciS 1
1−
+ciS c
iS 1−
28
19.03.2009 © SUPSI-DTI 55
Dipartimento
Tecnologie
Innovative
Variation of the sets of encounternodes
�The variation of the set of encounter nodes in a period is indicated by
� is computed with respect to the past cycle (same period) and to the previous period (samecycle)
iEV
iEV
ci
ci
ci
ci
ci
ci
SS
SSSS
11
11
11
−−
−−
−−
∪
∩−∪=
11−
+c
iT1−ciT c
iT 1−
1−ciS 1
1−
+ciS c
iS 1−
iEV
iEV
19.03.2009 © SUPSI-DTI 56
Dipartimento
Tecnologie
Innovative
SpatioTempo – Design
�Each person is supposed to have a set offrequent contact persons (FD).
�The source node S wants to send a message M to the destination D. There are 2 cases:� D belongs to FD
� D is an occasional destination (D belongs to OD)
29
19.03.2009 © SUPSI-DTI 57
Dipartimento
Tecnologie
Innovative
Recording task
11−
+c
iT1−ciT c
iT 1−
Slidingwindowcycle
1−ciS
11−
+ciS
1−ciS
11−
+ciS
1−ciS
12
−+ciS
ciS 1−
11−
+ciS
1−ciS
19.03.2009 © SUPSI-DTI 58
Dipartimento
Tecnologie
Innovative
Compute the number of retransmissions
�D belongs to OD: Basically use Propicmanto select the next hop with the number (F) of rebroadcasting , depends on thebattery level (b), the priority level ofmessage ( ) and in each period.
0 if b < (0< <1)F=
round( * * ): number of units of time in the period
iEV
HM
iEV
µ
ββ
µit
it
30
19.03.2009 © SUPSI-DTI 59
Dipartimento
Tecnologie
Innovative
D belongs to FD� Delivery probability of a period:
Delivery probability to reach the destination if the sendersends the message in this period.
� The number of retransmissions:
F= round( )
, : maximum and minimum number of times that S isable to send during the periodmF
1*
*)(**
++−+
ic
i
mc
imMic
iM
EVDP
FDPFFEVDPF
ciDP
ciDP
ηη+
+= −−
1
* 11 c
ici DPDP
MF
19.03.2009 © SUPSI-DTI 60
Dipartimento
Tecnologie
Innovative
D belongs to FD�Given is number of units of time in the period
�And if S can resend each unit of time in theperiod, then equals
F = round ( ) 1*
1*)1(**
++−+
ic
i
cii
cii
EVDP
DPtEVDPt
MF itHM
it
31
19.03.2009 © SUPSI-DTI 61
Dipartimento
Tecnologie
Innovative
11−
+c
iT1−ciT
Cycle
ciT 1−
S S S S S SHM
HM
HMHM
HM
HM
HM
DP
DP DP
DP
maxDP
maxDP
maxDP
19.03.2009 © SUPSI-DTI 62
Dipartimento
Tecnologie
Innovative
OD/FD management & Message lifetime� After one cycle, nodes review its OD, FD lists� Each member in FD, OD of S has a frequency level (FL), whichindicates how often S has a message to it.
� Message lifetime:
: The duration of time of a cycle
TTTL CM *µ=
OD FD
: demotedδ>DFL
: promotedδ>DFL: removedδ>DFL
TC
32
19.03.2009 © SUPSI-DTI 63
Dipartimento
Tecnologie
Innovative
Predict next best interval
�Having information on the past cycle and past interval, we can predict next best interval
11−
+c
iT1−ciT c
iT 1−
1−ciS 1
1−
+ciS c
iS 1−
ciDP
19.03.2009 © SUPSI-DTI 64
Dipartimento
Tecnologie
Innovative
Opportunistic Sensor Networks� Context oblivious: Data Mules, SWIM, ZebraNet, Message Ferrying, DFT-MSN
� Z. Zhang. Routing in intermittently connected mobile ad hoc networks and delay tolerant networks: overview and challenges. IEEE Communications Surveys & Tutorials, 2006.
� R. C. Shah, S. Roy, S. Jain, W. Brunette. Data MULEs: Modeling a three-tier architecture for sparse sensor networks. SNPA, 2003.
� Muhammad Mukarram Bin Tariq, Mostafa Ammar, Ellen Zegura. Message Ferry Route Design for Sparse Ad Hoc Networks with Mobile Nodes. In Proceedings of ACM Mobihoc06, 2006.
33
19.03.2009 © SUPSI-DTI 65
Dipartimento
Tecnologie
Innovative
Opportunistic Sensor Networks
�Context aware� SCAR = Sensor based Context-Aware Routing
�Opportunistic Mobile Sensor Data Collection with SCAR. Bence Pasztor, Mirco Musolesi and Cecilia Mascolo. In Proceedings of MASS 2007
� PROSAN: PRobabilistic Opportunistic Routing in SANets
� Probabilistic Opportunistic Routing in SANETs. A.H. Nguyen, S. Giordano. In Proceedings of ACM SANET – MobiCOM2007.
19.03.2009 © SUPSI-DTI 66
Dipartimento
Tecnologie
Innovative
Motivation�Pervasiveness and repeating behavior are true also in SANET:
� people who repeatedly stay and pass in area A are the ones who most likely want, need and will support a reaction.
�By naturally acting, people can enable the communication between sensors and actuators, and trigger the reaction.
34
19.03.2009 © SUPSI-DTI 67
Dipartimento
Tecnologie
Innovative
Motivating Scenario
�same for animals
19.03.2009 © SUPSI-DTI 68
Dipartimento
Tecnologie
Innovative
SCAR�Mobile sensors collect information and try to deliver this opportunistically to one of the sinks� Energy is an issue (batteries cannot be changed so easily)
�They exploit “best” carriers to deliver the information back to sinks� Choice of the best carriers is key to the protocol: SCAR uses utility theory and Kalman Filter based prediction of various attributes to choose
35
19.03.2009 © SUPSI-DTI 69
Dipartimento
Tecnologie
Innovative
Assumptions
�Data is reasonably delay tolerant
�Clocks of the sensors are (reasonably) synchronized and the sensors wake up at regular intervals
�Sensors are all cooperating (possibly owned by same developer)
�Mobility is not source of energy consumption
19.03.2009 © SUPSI-DTI 70
Dipartimento
Tecnologie
Innovative
The key ideas of SCAR�Each sensor collects and stores its data
�Each sensor will try to deliver its data in bundles to a number of neighbouring sensors which seem to be the best carriers to reach a sink
�Decision process based on the prediction of the evolution of the system:
� mobility of nodes
� colocation with the sinks
� battery level
�The forecasted values of the context attributes are then combined to define a probability P(si) of the sensorsi of delivering the bundles to one of the sinks
36
19.03.2009 © SUPSI-DTI 71
Dipartimento
Tecnologie
Innovative
Multiple-carrier Selection�Each node maintains a list of the neighbouring nodes (including itself) ordered according to their delivery probabilities
�Each source node then replicates the bundle to the first R nodes (R-1 nodes if the node itself is in the first Rpositions of the list)
� Data bundles are only replicated on R nodes by the source, while they are forwarded (deleted from a node and copied on another) if the carriers, while roaming, find either a sink or abetter carrier.
� The value of R is specified by the user and can be considered as a priority level associated to the data retrieved by the sensor
19.03.2009 © SUPSI-DTI 72
Dipartimento
Tecnologie
Innovative
Bundle Forwarding
�Each node keeps monitoring if there are neighbours with better probability of delivery than its own
�If this is the case the data bundles are shifted to the other nodes
�Key problem: intelligent replication
37
19.03.2009 © SUPSI-DTI 73
Dipartimento
Tecnologie
Innovative
Forecasting Techniques for Probabilistic Routing
�Each node calculates its delivery probability using multi-criteria decision theory over the evolution of its context described by a set of attributes.
�In particular, we consider three attributes:� Colocation with the sinks� Change degree of connectivity (ie how many nodes have been
seen in a fraction of time)� Battery level
�Our aim is to maximize the combined utility function(i.e., selection of the nodes with the best trade-off between these attributes)
19.03.2009 © SUPSI-DTI 74
Dipartimento
Tecnologie
Innovative
Prosan design�Based on user profiles
� Using users’ social behavior – node profile - to predict the mobility of nodes.
�Routing with Zero Knowledge� Sender does not need to know any information about other nodes – no synchronizing, no local table.
� Only senders broadcast helloing messages.
�No sharing user information� Nodes does not need to share, periodically update their information to others.
� Keep privacy amongst members.
38
19.03.2009 © SUPSI-DTI 75
Dipartimento
Tecnologie
Innovative
Forwarding strategy�Delivery probability (DP): The probability of a node to meet and forward a message to other nodes toward destination.
�The sender S wants to send a message M to the destination actuator D, S will choose the route that has the highest delivery probability (DP) to send M to D.
�Before sending the message content M to the destination D, S sends M’s header to its neighbors.
hM = (H(EDi,VDi), MACS, SNM) (1)Example:26a48976c47c0d2b2383a34010eae937ebcf8d1f6c554b9f7214a12de02cd9c0c5fd9486e7967327b60f6d2b008ff1c110c09b62a536a195c27b76b17564e92f25260a3c6599f05fef87f3110b7f171871aef40a2b17fc1bbf88053aacea724c0a08419a9f211b077bcc244c056817c8c4ca4238a0b923820dcc509a6f75849b
n
iionConcatenat
1=
19.03.2009 © SUPSI-DTI 76
Dipartimento
Tecnologie
Innovative
Forwarding strategy (cont)
Sender
AhM
……
B
hMhMhMhMhMhMhM
H(En,Vn)VnEn
………
H(E2,V2)V2E2
H(E1,V1)V1E1
Hashed values
Values
Evidence names
Node profile
H(ED1,VD1)
H(ED2,VD2)
H(ED3,VD3)
H(ED4,VD4)
H(EDn,VDn)
MACS SN
Message header hM …
Hashed values in node profile
H(E1,V1)
H(E2,V2)
H(E3,V3)
H(E4,V4)
H(En,Vn)
…
39
19.03.2009 © SUPSI-DTI 77
Dipartimento
Tecnologie
Innovative
Forwarding strategy (cont)� Neighbors do local matching between received hM with its node profile.
� With each matching, neighbor nodes look up in T to get the corresponding weight. Then neighbor node (A) can compute its DPAas follows:
0≤ DPA = ≤1
� : sum of the weights of all the evidences that S knows about D.
H(ED1,VD1)
H(ED2,VD2)
H(ED3,VD3)
H(ED4,VD4)
H(EDn,VDn)
MACS SN
Message header hM
…
Hashed values in node profile
H(E1,V1) H(E2,V2) H(E3,V3) H(E4,V4) H(En,Vn)…
∑∑
Mi
i
W
Wmatched )(
∑ iMW
19.03.2009 © SUPSI-DTI 78
Dipartimento
Tecnologie
Innovative
Forwarding strategy (cont)
Sender
AhM
……
B
hMhMhMhMhMhMhM
0≤ DPN = ≤1
DP
DP
DPDP
DP
DP
DP
∑∑
Mi
i
W
WMatched )(
Delivery Probability DP
� S will choose the route(s) with highest DP it receives from its neighbor nodes
� S only sends the message to these routes if DProute > DPS
40
19.03.2009 © SUPSI-DTI 79
Dipartimento
Tecnologie
Innovative
Simulation environment� Hybrid P-MULEs
� nodes with higher capacity� use Prosan� move as MULEs
� Simulation area� rectangle 870x520 meters� divided in 6 areas� 50 nodes (1 p-mule)� use sleep mode
� Simulation approach� static actuator� mobile actuator
trajectory of p-mule
19.03.2009 © SUPSI-DTI 80
Dipartimento
Tecnologie
Innovative
Overhead
0200400600800
10001200140016001800
1 30
Tests
Byt
es
P-Mules
PROSAN
Hybrid P-Mules
Delay
0
10000
20000
30000
40000
50000
1 30
Tests
Mse
c P-Mules
PROSAN
Hybrid P-Mules
Simulation results for static actuator
41
19.03.2009 © SUPSI-DTI 81
Dipartimento
Tecnologie
Innovative
Overhead
0
200
400
600
800
1000
1200
1400
1 30
Tests
Byt
es
P-Mules
PROSAN
Hybrid P-Mules
Delay
0
10000
20000
30000
40000
50000
60000
70000
1 30
Tests
Mse
c
P-Mules
PROSAN
Hybrid P-Mules
Simulation results for mobile actuator
19.03.2009 © SUPSI-DTI 82
Dipartimento
Tecnologie
Innovative
Conclusion�context information in opportunistic routing can be very advantageous
�exploit effectively the mobility, the social connections and the community relationships:
� reduce significantly the number of nodes involved in� reduce the number of broadcasts� efficiently control message lifetime
�assumption for effectively using context information� people move in a predictable fashion based on repeating behavioral
patterns at different timescales (day, week, and month) � or these patterns are known
� -> mobility as expression of the users social behavior�Haggle architecture for supporting context aware routing
42
19.03.2009 © SUPSI-DTI 83
Dipartimento
Tecnologie
Innovative
Thanks