TRANSFER: Transactions Routing for Adhoc NetworkS with eFficient EneRgy

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 helmy_at_ufl.edu
• web www.cise.ufl.edu/helmy
• Wireless Networking Lab nile.cise.ufl.edu

2
Motivation

• Most current ad hoc routing approaches
• Setup/maintain optimal (e.g., shortest) routes
  (DSR, AODV, ZRP,..)
• Incur high route discovery cost, warranted for
  long-lived flows where cost is amortized over
  flow duration
• In Small Transactions
• Cost is dominated by route discovery (vs. data
  transfer)
• Design Goal reduce cost for small transactions
• Example small transactions resource discovery
  query, text messaging, sensor network query, etc.

3
Approach

• Avoid flooding-based approaches and instead of
  flat architecture use hierarchical architecture
• Instead of complex hierarchy formation use loose
  hierarchy (zone-based)
• Instead of bordercasting (as in ZRP) query only a
  few selected contact nodes
• Contacts act as short cuts to bridge zones and
  reduce degrees of separation between querier
  resource
• Borrows from the concept of small worlds

4
Flooding vs. Contact-based Query
(a) Flooding from source (S) to discover Target
(b) Query from source (S) using contacts C1 and
C2 to discover Target
5
Architectural Overview
NoC Number of Contacts
6
Contact Selection Scheme

• Reactive (on-demand) contact selection
• Choose contacts with reduced proximity overlap
• 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

7
Overlap Problem and Solution
B avoids going through Ls neighbors x, y,
z (Straightening algorithm)
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

9
Single-shot Policy
NoC3 D2 R3 r3
10
Level-by-level or Step Policy
contact-2
NoC3 D2 R3 r3
11
Attempt 3
Attempt 2
Attempt 2
Attempt 3
Q
Attempt 2
Attempt 3
Rotation-like effect in the step search policy
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

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)
• EbEtxg.ErxEtx(1f.g), where g is ave. node
  degree.
• Energy consumed by a unicast message (Eu)
• EuEtxErxEhEtx(1fh), where fErx/Etx and
  hEh/Etx, Eh energy consumption due to handshake.
• For this study we use f0.64, and h0.1

14
Simulation Setup

• Random node layout
• Random way point mobility model 0,20 m/s
• Random src-dst pair selection
• R3 to limit storage and proximity overhead

15
Optimum Number of Contacts (NoC)
Reduced coverage frequent fallback to flooding
N1000 nodes
Increased query threads
, r3, D33 (5 attempts max)
- Optimum NoC3, resulting in (near) perfect
coverage
16
Optimum contact distance (r)
N1000 nodes
, NoC3, D33 (5 attempts max)
- Optimum r3, resulting in min overlap and max
coverage
17
Optimum depth of search (D)
2 attempts
3 attempts
N1000 nodes
4 attempts
5 attempts
, NoC3, r3
- D33 (5 attempts max) results in (near) perfect
coverage - High order attempts (4th 5th) only
search unvisited parts of the network (due to
re-visit prevention) and achieve increased
coverage without excessive overhead
18
Scalability Analysis and Comparisons
(1) Per-Query Energy Consumption
(NoC3, r3, D33)
- Total query energy consumption f(query rate)
query/s - Define per-query energy as Estep,
Eflood and Eborder
19
Comparisons (contd.)
(2) Proximity (Zone) Maintenance Energy
Consumption
- For TRANSFER Z(R)Z(3), for ZRP Z(2R-1)Z(5)
(extended zone) - Proximity costf(mobility) m/s
20
Comparisons (contd.)
Total Energy Consumption Proximity Query Energy

• 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 ETstepZ(R)q.Estep
• Total Flood Energy ETfloodq.Eflood
• Total ZRP Energy ETborderZ(2R-1)q.Eborder
• Define total energy ratios (TER)

21
Comparisons (contd.)
(3.a) Total Energy Consumption (vs. Flooding)
- For high query rates achieves energy savings of
90-95 over flooding
22
Comparisons (contd.)
(3.b) Total Energy Consumption (vs. ZRP
bordercasting)
- For high query rates achieves energy savings of
75-86 over ZRP
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

24
Backup Slides
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
Query Resolution Latency
- For single-shot minimum number of attempts
(1) - For step number of attempts scales well
with network size
29
Comparisons
ODC on-demand routing with caching
(DSR-like) MDS minimum dominating set
algorithm Smart-fld smart flooding
(location-based heuristic)
30
Comparisons
ODC on-demand routing with caching
(DSR-like) MDS minimum dominating set
algorithm Smart-fld smart flooding
(location-based heuristic)