Ahmed Helmy - UFL 1 TRANSFER: Transactions Routing for Ad-hoc NetworkS with eFficient EneRgy Ahmed Helmy Computer and Information Science and Engineering (CISE) University of Florida (UFL) email: [email protected]web: www.cise.ufl.edu/~helmy Wireless Networking Lab: nile.cise.ufl.edu
30
Embed
TRANSFER : Transactions Routing for Ad-hoc NetworkS with eFficient EneRgy
TRANSFER : Transactions Routing for Ad-hoc NetworkS with eFficient EneRgy. Ahmed Helmy Computer and Information Science and Engineering (CISE) University of Florida (UFL) email: [email protected] web: www.cise.ufl.edu/~helmy Wireless Networking Lab: nile.cise.ufl.edu. Motivation. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Ahmed Helmy - UFL 1
TRANSFER: Transactions Routing for Ad-hoc NetworkS with eFficient EneRgy
Ahmed HelmyComputer and Information Science and Engineering (CISE)
• Proximity overlap reduction mechanisms– use the proximity information at the border (if
available as link state) to reduce the overlaps– use the neighbor-neighbor avoidance mechanism– use disjoint paths (as possible) to reach contacts
Ahmed Helmy - UFL 7
RR
Q
contact
Q: Querier NodeB: Border Node
C: Contact NodeR: zone radius
B
C1
R
xL
R
R
Q
contact
tr: transmission trange
tr
L
BC2
R
x
y
z
B avoids going through L’s neighbors x, y, z(Straightening algorithm)
Overlap Problem and Solution
Ahmed Helmy - UFL 8
Search Policies• Levels of contacts defined by maximum depth D
• Several search policies investigated:– Single-shot uses 1 attempt (minimum latency)– Level-by-level uses several attempts with depth level
increased by 1 for every attempt– Step uses several attempts with depth increased
exponentially 1,2,4,8,… (minimum overhead)
• In multi-attempts use the rotation effect– choose different level-1 contacts for different attempts to
increase network coverage
• Use loop detection and re-visit prevention
Ahmed Helmy - UFL 9
Q
contact-1
contact-1
contact-1
contact-2contact-2
contact-2
contact-2
contact-2
contact-2
contact-2
contact-2
contact-2
Single-shot Policy
NoC=3D=2R=3r=3
Ahmed Helmy - UFL 10
contact-1
contact-1
contact-1
Qcontact-2
contact-1
contact-2
contact-2
contact-1
contact-2contact-2
contact-2
contact-1
contact-2
contact-2
contact-2
Level-by-levelor Step Policy
NoC=3D=2R=3r=3
Ahmed Helmy - UFL 11
Attempt 1
Attempt 1
Attempt 1
Attempt 2
Attempt 2
Attempt 2
Attempt 3
Attempt 3
Attempt 3
Q
Rotation-like effect in the step search policy
Ahmed Helmy - UFL 12
Evaluation and Analysis
• Trade-off between success rate vs. energy
• Simulation uses fallback to flooding upon failure
• Parameter analysis (optimum r, NoC, D)
• Main evaluation metric is total energy consumption
• Energy consumption due to various components– Proximity maintenance: function of mobility m/s– Query overhead: function of query rate query/s– Total Consumption: function of q (query/s)/(m/s) QMR
Ahmed Helmy - UFL 13
The Communication Energy Model Based on IEEE 802.11
Accounts for energy consumption due to transmission and reception
Accounts for differences between broadcast and unicast messages
Energy consumed by a broadcast message (Eb):– Eb=Etx+g.Erx=Etx(1+f.g), where g is ave. node degree.
Energy consumed by a unicast message (Eu):– Eu=Etx+Erx+Eh=Etx(1+f+h), where f=Erx/Etx and h=Eh/Etx, Eh energy
- Optimum NoC=3, resulting in (near) perfect coverage
Ahmed Helmy - UFL 16
Optimum contact distance (r)
, NoC=3, D=33 (5 attempts max)
N=1000 nodes
- Optimum r=3, resulting in min overlap and max coverage
Ahmed Helmy - UFL 17
Optimum depth of search (D)
3 attempts
2 attempts
4 attempts 5 attempts
- D=33 (5 attempts max) results in (near) perfect coverage- High order attempts (4th & 5th) only search unvisited partsof the network (due to re-visit prevention) and achieve increased coverage without excessive overhead
, NoC=3, r=3
N=1000 nodes
Ahmed Helmy - UFL 18
(1) Per-Query Energy Consumption
Scalability Analysis and Comparisons
(NoC=3, r=3, D=33)- Total query energy consumption = f(query rate) query/s- Define per-query energy as Estep, Eflood and Eborder
Ahmed Helmy - UFL 19
0
50
100
150
200
250
200 500 1000 2000 4000 8000 16000 32000
Network Size (nodes)
Ene
rgy
pe
r n
od
e p
er
se
c p
er
m/s Z(3)
Z(5)
(2) Proximity (Zone) Maintenance Energy Consumption
Comparisons (contd.)
- For TRANSFER Z(R)=Z(3), for ZRP Z(2R-1)=Z(5) (extended zone)- Proximity cost=f(mobility) m/s
Ahmed Helmy - UFL 20
Comparisons (contd.)
– To combine the proximity energy, f(mobility), and the query energy, f(query rate)
– The query-mobility-ratio (QMR) metric, q, in query/s/(m/s) is used for normalization
• Total Step Energy: ETstep=Z(R)+q.Estep
• Total Flood Energy: ETflood=q.Eflood
• Total ZRP Energy: ETborder=Z(2R-1)+q.Eborder
– Define total energy ratios (TER):
flood
step
Tflood
Tstepflood Eq
EqRZ
E
ETER
.
.)(
border
step
Tborder
Tstepborder EqRZ
EqRZ
E
ETER
.)12(
.)(
Total Energy Consumption: Proximity + Query Energy
Ahmed Helmy - UFL 21
(3.a) Total Energy Consumption (vs. Flooding)
Comparisons (contd.)
- For high query rates achieves energy savings of 90-95% over flooding
Ahmed Helmy - UFL 22
(3.b) Total Energy Consumption (vs. ZRP bordercasting)
Comparisons (contd.)
- For high query rates achieves energy savings of 75-86% over ZRP
Ahmed Helmy - UFL 23
Summary/ Conclusions• Developed a contact-based architecture for
energy-efficient routing of small transactions
• Introduced effective contact selection scheme
• Investigated several search policies (e.g., Step)
• Analyzed performance of TRANSFER and showed favorable parameter settings for a wide array of networks
• Achieved gains for high query rates 75-95% as compared to flooding and ZRP
Ahmed Helmy - UFL 24
Backup Slides
Ahmed Helmy - UFL 25
500
700
900
1100
1300
1500
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Contact Distance (r )
Ene
rgy
Con
sum
ed p
er q
uery
(Etx
uni
ts) lbl
step
single-shot
Ahmed Helmy - UFL 26
500
600
700
800
900
1000
1100
1200
1300
2 3 4 5 6 7 8Number of Contacts (NoC )
Ene
rgy
per
quer
y (E
tx u
nits
)
step
lbl8
single-shot
Ahmed Helmy - UFL 27
0
1000
2000
3000
4000
5000
1 6 11 16 21 26 31 36
maxDepth (D )
En
erg
y p
er q
uer
y (E
tx u
nit
s)
step
lbl
single-shot
Ahmed Helmy - UFL 28
0
2
4
6
8
10
12
14
200 500 1000 2000 4000 8000 16000 32000
Network size, N (nodes)
Ave
rage
num
ber
of a
ttem
pts lbl
step
single-shot
Query Resolution Latency
- For single-shot: minimum number of attempts (~1)- For step: number of attempts scales well with network size
Ahmed Helmy - UFL 29
0
50000
100000
150000
200000
250000
300000
0 5000 10000 15000 20000 25000 30000 35000
Network Size (nodes)
Ene
rgy
per
quer
y (E
tx u
nits
)
FloodingODCSmart-fldMDSZRPlblStep
Comparisons
ODC: on-demand routing with caching (DSR-like)MDS: minimum dominating set algorithmSmart-fld: smart flooding (location-based heuristic)
Ahmed Helmy - UFL 30
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
200 500 1000 2000 4000 8000 16000 32000
Network size, N (nodes)
Que
ry E
nerg
y R
atio
(vs.
Flo
od)
ODC/FldSmartFld/FldMDS/FldZRP/FldStep/Fld
Comparisons
ODC: on-demand routing with caching (DSR-like)MDS: minimum dominating set algorithmSmart-fld: smart flooding (location-based heuristic)