Two Hops or More:
On Hop Limited Search in Opportunistic Networks
Suzan Bayhan† , Esa Hyytiä‡ , Jussi Kangasharju† and Jörg Ott‡,?
University of Helsinki, Finland † , Aalto University, Finland ‡ , and Technical University of Munich, Germany?
MSWiM 2015
1/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Searching content in opportunistic networks
• Useful information often available locally due to spatial locality,
homophily, etc. but may not be accessible using the convential
techniques (i.e., Internet)
2/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Searching content in opportunistic networks
• Useful information often available locally due to spatial locality,
homophily, etc. but may not be accessible using the convential
techniques (i.e., Internet)
• Find content using opportunistic search
2/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Searching content in opportunistic networks
• Useful information often available locally due to spatial locality,
homophily, etc. but may not be accessible using the convential
techniques (i.e., Internet)
• Find content using opportunistic search
• No Google-like service!
2/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Searching content in opportunistic networks
• Useful information often available locally due to spatial locality,
homophily, etc. but may not be accessible using the convential
techniques (i.e., Internet)
• Find content using opportunistic search
• No Google-like service! ! Multi-copy multi-hop routing (increased
redundancy)
• Previous research: publish/subscribe (push approach)
content
Consumer
Provider
subscribe
2/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Searching content in opportunistic networks
• Useful information often available locally due to spatial locality,
homophily, etc. but may not be accessible using the convential
techniques (i.e., Internet)
• Find content using opportunistic search
• No Google-like service! ! Multi-copy multi-hop routing (increased
redundancy)
• Previous research: publish/subscribe (push approach)
content
Consumer
Provider
subscribe
• Pull: more natural like desktop search, receiver driven
query
Provider
Consumer
content 2/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Hop-limited search
Search starts
3/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Hop-limited search
Search starts
Content provider
receives the query
Multi-hop multi-copy routing
4/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Hop-limited search
Search starts
Content provider
receives the query
Content reaches
the user
Multi-hop multi-copy
routing
Multi-hop multi-copy routing
5/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Hop-limited search
Search starts
Content provider
receives the query
Forward path
Content reaches
the user
Return path
Multi-hop multi-copy
routing
Multi-hop multi-copy routing
6/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
How many hops to go?
We want to understand the effect of hop count on:
• search success ratio, search completion time, search cost
7/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
How many hops to go?
We want to understand the effect of hop count on:
• search success ratio, search completion time, search cost
• Decision on hop count is not clear in the literature: two hops or
more?
7/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Three components of search
1
Content: scarce item or densely available item?
2
User’s tolerated waiting time: low patience or tolerant to delay?
3
Search cost: avoid bandwidth waste and battery consumption
8/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Three components of search
1
Content: scarce item or densely available item?
•
2
User’s tolerated waiting time: low patience or tolerant to delay?
•
3
Content availability: ↵
TTL of the message: T
Search cost: avoid bandwidth waste and battery consumption
•
hop-limit: h
8/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Modeling of hop-limited search
• Total N nodes in the network
• Content availability ↵
• Assume only K messages are allowed for forward path
9/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Modeling of hop-limited search
• Total N nodes in the network
• Content availability ↵
• Assume only K messages are allowed for forward path
Ps =
K
X
P r{k content providers are discovered}
k=1
⇥ P r{at least one of k responses reaches searching node}.
9/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Modeling of hop-limited search
• Total N nodes in the network
• Content availability ↵
• Assume only K messages are allowed for forward path
Ps =
K
X
P r{k content providers are discovered}
k=1
⇥ P r{at least one of k responses reaches searching node}.
Ps = 1
(1
↵ )K where
9/29
=
K
N
1
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
• Content availability ↵ (available or
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
estimated)
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
• Content availability ↵ (available or
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
estimated)
•
= NK 1
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
• Content availability ↵ (available or
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
estimated)
•
= NK 1
• K: f (hop limit, message lifetime)! Nh (T )
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
• Content availability ↵ (available or
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
estimated)
•
= NK 1
• K: f (hop limit, message lifetime)! Nh (T )
• Nh (T ):
•
•
for static networks, see Wang et. al ICN 2015
difficult to model realistically for mobile
networks
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success
(1
↵ )K
• Content availability ↵ (available or
0.8
0.4
0.0
Success ratio, Ps
Ps = 1
Forward, 0.05
Forward, 0.15
Forward, 0.4
Complete search, 0.05
Complete search, 0.15
Complete search, 0.4
0.05 0.25 0.45 0.65 0.85
Frac.of nodes receiving msg., K/(N−1)
estimated)
•
= NK 1
• K: f (hop limit, message lifetime)! Nh (T )
• Nh (T ):
•
•
for static networks, see Wang et. al ICN 2015
difficult to model realistically for mobile
networks
Our approach: find Nh (T ) from real traces and plug into Ps
10/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success ratio: simulations
• Infocom06 (98 nodes, conference)
• Cabspotting (around 500 nodes, San Francisco cabs)
• Helsinki City Scenario (synthetic)
11/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search success ratio: simulations
• Infocom06 (98 nodes, conference)( Nh (T ) and query success
• Cabspotting (around 500 nodes, San Francisco cabs)
T=600 s
T=2600 s
T=4600 s
T=6600 s
0.70
T=600 s
T=2600 s
T=4600 s
T=6600 s
Forward success ratio
0.80
0.90
Neighborhood size
0.3 0.4 0.5 0.6 0.7 0.8 0.9
1.00
• Helsinki City Scenario (synthetic)
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
11/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
T=600 s
T=2600 s
T=4600 s
T=6600 s
0.70
Key take-away:
• Second hop brings the highest
benefits for content discovery
• But still improvements for h 6 4
• Minimal improvement ! not
significant increase in user
satisfaction
Forward success ratio
0.80
0.90
1.00
Search success ratio: simulations
1
12/29
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search completion time
Continuous Time Markov Chain modeling of the forward path: Th
• Return path = forward path with
↵=
Tagged
1
Meeting a
tagged node
N 1
• T̃h : an approximation for Th ,
meeting rate , details in the
paper!
T̃h =
1
X
i=0
Meeting a
tagged node
Start
0-hop
(1,1,0,%,0)
Meeting a
tagged node
Meeting an
undiscovered node
g
meetin
i-hop m
eeting
i
(1 ↵/ )
.
(1 + i(1 h 1 ))
Meeting a
tagged node
(i,j)-meeting
Subset of states
Sm = (m, 1, . . . , mh )
13/29
(h1)h
op
m
ee
tin
g
Subset of states
Sm+1 = (m + 1, 1, . . . , mh )
Subset of states
SN
M
= (N
M, 1, . . . , mh )
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search completion time: CTMC model
↵
Low
Medium
High
Time
Th
T̃h
Error
Th
T̃h
Error
Th
T̃h
Error
h=1
1
1
0
1
1
0
1
1
0
h=2
0.21
0.14
-0.32
0.40
0.39
-0.01
0.51
0.54
0.05
h=3
0.13
0.12
-0.08
0.33
0.35
0.07
0.46
0.49
0.08
h=4
0.12
0.11
-0.03
0.31
0.33
0.06
0.45
0.47
0.06
h=5
0.11
0.11
-0.03
0.31
0.33
0.05
0.45
0.46
0.04
• Notice the drastic decrease at h = 2
14/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search completion time: CTMC model
↵
Low
Medium
High
Time
Th
T̃h
Error
Th
T̃h
Error
Th
T̃h
Error
h=1
1
1
0
1
1
0
1
1
0
h=2
0.21
0.14
-0.32
0.40
0.39
-0.01
0.51
0.54
0.05
h=3
0.13
0.12
-0.08
0.33
0.35
0.07
0.46
0.49
0.08
h=4
0.12
0.11
-0.03
0.31
0.33
0.06
0.45
0.47
0.06
h=5
0.11
0.11
-0.03
0.31
0.33
0.05
0.45
0.46
0.04
• Second hop brings the highest gain!
15/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search completion time: CTMC model
↵
Low
Medium
High
Time
Th
T̃h
Error
Th
T̃h
Error
Th
T̃h
Error
h=1
1
1
0
1
1
0
1
1
0
h=2
0.21
0.14
-0.32
0.40
0.39
-0.01
0.51
0.54
0.05
h=3
0.13
0.12
-0.08
0.33
0.35
0.07
0.46
0.49
0.08
h=4
0.12
0.11
-0.03
0.31
0.33
0.06
0.45
0.47
0.06
h=5
0.11
0.11
-0.03
0.31
0.33
0.05
0.45
0.46
0.04
• T̃h provides pretty accurate approximation of Th
16/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
How about in the wild?
17/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Complete search: Simulations
• Realism included: simulations using ONE simulator
• We want to understand:
• validity of our conclusions from the model (second hop!)
• average (temporal and hop) distance to content provider
• average (temporal and hop) distance to the searching node
• forward path vs. return path
• search cost
18/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
(a) Infocom06
1.0
Search success
0.4
0.6
0.8
1.0
T=3600
T=21600
T=86400
0.2
T=3600
T=21600
T=86400
0.2
0.2
T=3600
T=21600
T=86400
Search success
0.4
0.6
0.8
Forward path success ratio
0.4
0.6
0.8
1.0
Search success: low content availability (↵ = 0.05)
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
(b) Infocom06
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
SF cabs
• Second hop!
• High tolerated waiting time: wait till meeting the content provider!
• SF cabs have higher contact opportunities
19/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Where is the content?
No hop or time restriction, epidemic search
20/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Where is the content?
0
0
Hop distance
2 4 6 8
Cabspotting
Hop distance
2 4 6 8
Infocom06
Medium
Content availability
Query
Low
Medium
Content availability
High
Response
High
Low
Medium
Content availability
High
Low
Medium
Content availability
High
Temporal distance
0
400 800
Temporal distance
0 5000
15000
Low
Left bars: query, right bars: response
21/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
• Content’s location depends on content availability: forward path
• Distance to searching node: independent of content availability
0
0
Hop distance
2 4 6 8
Cabspotting
Hop distance
2 4 6 8
Infocom06
Medium
Content availability
Query
Low
Medium
Content availability
High
Response
High
Low
Medium
Content availability
High
Low
Medium
Content availability
High
Temporal distance
0
400 800
Temporal distance
0 5000
15000
Low
22/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Forward and return paths: are they correlated?
• ⇢: Pearson correlation coefficient of forward and return paths
•
: Difficulty of return path compared to forward path (i.e., ratio)
T
600
3600
86400
↵
Low
Med.
High
Low
Med.
High
Low
Med.
High
⇢hop
0.30
0.29
0.27
0.35
0.35
0.33
0.33
0.35
0.35
⇢temp
0.36
0.34
0.32
0.43
0.38
0.32
0.13
0.13
0.12
hop
temp
1.47
1.72
1.97
1.4
1.61
1.85
1.39
1.62
1.86
2.23
3.09
4.02
2.18
2.98
4.13
2.60
3.52
4.50
23/29
Ph
0.35
0.42
0.48
0.63
0.67
0.70
0.95
0.95
0.95
Ps
0.30
0.34
0.38
0.57
0.61
0.65
0.94
0.94
0.95
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
No strong correlation (effect of restricted tolerated time)
T
600
3600
86400
↵
Low
Med.
High
Low
Med.
High
Low
Med.
High
⇢hop
0.30
0.29
0.27
0.35
0.35
0.33
0.33
0.35
0.35
⇢temp
0.36
0.34
0.32
0.43
0.38
0.32
0.13
0.13
0.12
hop
temp
1.47
1.72
1.97
1.4
1.61
1.85
1.39
1.62
1.86
2.23
3.09
4.02
2.18
2.98
4.13
2.60
3.52
4.50
24/29
Ph
0.35
0.42
0.48
0.63
0.67
0.70
0.95
0.95
0.95
Ps
0.30
0.34
0.38
0.57
0.61
0.65
0.94
0.94
0.95
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Return path is more challenging
T
600
3600
86400
↵
Low
Med.
High
Low
Med.
High
Low
Med.
High
⇢hop
0.30
0.29
0.27
0.35
0.35
0.33
0.33
0.35
0.35
⇢temp
0.36
0.34
0.32
0.43
0.38
0.32
0.13
0.13
0.12
hop
temp
1.47
1.72
1.97
1.4
1.61
1.85
1.39
1.62
1.86
2.23
3.09
4.02
2.18
2.98
4.13
2.60
3.52
4.50
25/29
Ph
0.35
0.42
0.48
0.63
0.67
0.70
0.95
0.95
0.95
Ps
0.30
0.34
0.38
0.57
0.61
0.65
0.94
0.94
0.95
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search cost
• Ideally, once content is discovered or response is received, message
spreading should be stopped
• Difficult in a distributed setting
26/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search cost
• Ideally, once content is discovered or response is received, message
spreading should be stopped
• Difficult in a distributed setting
ORACLE
EXCH
State of all messages
LOCAL
State of all messages
26/29
State of only shared messages
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Search time (s)
600
650
ORACLE
EXCH
LOCAL
0.08
ORACLE
EXCH
LOCAL
550
Query spread ratio
0.10 0.12 0.14
0.16
700
Search cost with increasing hop count
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
1
2
3 4 5 6 7 8 9 10
Hop count limitation (h)
• Even local knowledge is sufficient, thanks to the small network
diameter!
• Two-hops improves a lot, but further hops help decreasing search
completion time.
27/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Take aways from our paper
• Hop-limited search:
• content availability and mobility
• tolerated waiting time
• Second hop brings the highest benefits but further hops still improve
search performance
• Small world networks, i.e., no benefit after 4-5 hops
• Return path is more challenging
Thanks!
http://www.netlab.tkk.fi/tutkimus/pdp/
supported by Finnish Academy of Science
28/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan
Related Works
• Optimal hop count: MsWiM 2014
Esa Hyytiä, Suzan Bayhan, Jörg Ott, Jussi Kangasharju
• Availability estimation: ComCom 2015
Esa Hyytiä, Suzan Bayhan, Jörg Ott, Jussi Kangasharju
29/29
MSWiM’15
November 2–6, 2015, Cancun, Mexico
suzan bayhan