Upload
katelyn-malone
View
30
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Scalable Data Aggregation for Dynamic Events in Sensor Networks. Kai-Wei Fan, Sha Liu, Prasun Sinha Computer Science and Engineering, Ohio State University ACM SenSys 2006. Outline. Introduction Structure-Less Aggregation Experiments and Simulation Conclusion. Introduction. - PowerPoint PPT Presentation
Citation preview
Scalable Data Aggregation for Dynamic Events in Sensor Networks
Kai-Wei Fan, Sha Liu, Prasun Sinha
Computer Science and Engineering, Ohio State University
ACM SenSys 2006
Introduction
Data Aggregation Communication cost is often larger than computation cost. Redundancy in raw data. Aggregate packets near sources to reduce transmission cost.
Prolong the lifetime.
Aggregation Approaches Static structure Dynamic structure Structure-free
Static Structure for Aggregation Routing on a pre-computed structure
Pros Low maintenance cost Good for unchanged traffic pattern
Cons Long stretch problem Unsuitable for event-based network
Sink
Dynamic Structure for Aggregation Create a structure dynamically
Pros Optimization for source nodes
Cons High maintenance cost
Sink
Structure-Free Aggregation
No structure No structure maintenance cost
Aggregation without structure Where to transmit? Wait for whom?
Improve aggregating by transmitting packets to the same node at the same time Spatial Convergence Data Aware Anycast Temporal Convergence Random Waiting
Data Aware Anycast
Anycast One-to-any forwarding
Anycast to neighbor having packets for aggregating Class A: Nodes closer to the sink with data for aggregation Class B: Nodes with data for aggregation Class C: Nodes closer to the sink
Class B
Canceled CTS
Canceled CTS
RTS
CTS
Sender
Class A Nbr
Class B Nbr
Class C Nbr
Class A Nbr
Class A Class C
Random Waiting
Fixed Delay Nodes close to sink pick high delay.
Random Delay Source nodes pick random delay between 0 and τ before
transmission.
…Sink
τ=n τ=n-1 τ=n-2 τ=1 τ=0
Structure-Less Aggregation
Structure-free aggregation does not guarantee all packets are completely aggregated to one. High cost for un-aggregated or partial-aggregated packets
Structure-Less Aggregation (2 Phases) 1st : Based on structure-free aggregation (DAA & RW)
Aggregate packets form sources to aggregators locally
2nd : Further aggregation on an implicitly constructed structure Aggregate packets from aggregators to sink Tree on Directed Acyclic Graphic (ToD)
Tree on Directed Acyclic Graphic(ToD) Definition
Contiguous events Cell: A square area with side length greater than the diameter which an
event can span F-cluster: First cluster, composed of multiple cells S-cluster: Second cluster, composed of multiple cells (interleaved with F-
cluster)
1D Construction of ToD
F-cluster S-cluster
Tree on Directed Acyclic Graphic(ToD)
sink
F-clusters
F-cluster-head
Shortest Path
a b c d
F6
sink
S-cluster
S-cluster-headShortest Path
a b c d
S5 S6
sink
a b c d
Shortest Path Tree
F6
S6S5
Dynamic Forwarding for 1D (1) Forwarding Rules
Rule 0: Forward packets to F-aggregator by structure-free data aggregation protocol.
Rule 1: Event spans two cells in a F-cluster, forward to sink
Rule 2: Event spans one cells, forward to appropriate S-aggregator
sink
sink
Dynamic Forwarding for 1D (2) Property 1. Packets will be aggregated at a F-
aggregator, or will be aggregated at a S-aggregator.
If only nodes in one cell are triggered and generate the packets Aggregated at one F-aggregator (all nodes in a cell resides in the
same F-cluster)
If nodes in two cells are triggered and generate the packets. Two cells are in the same F-cluster
aggregated at the F-aggregator Two cells are in different F-clusters
aggregated at the S-aggregator
Tree on Directed Acyclic Grahpic(ToD) 2D Construction
A1 A2 B1 B2 C1 C2
A3 A4 B3 B4 C3 C4
D1 D2 E1 E2 F1 F2
D3 D4 E3 E4 F3 F4
G1 G2 H1 H2 I1 I2
G3 G4 H3 H4 I3 I4
A B
D E
G H
F
I
C
(a) F-clusters (b) Cells
A1 A2 B1 B2 C1 C2
A3 A4 B3 B4 C3 C4
D1 D2 E1 E2 F1 F2
D3 D4 E3 E4 F3 F4
G1 G2 H1 H2 I1 I2
G3 G4 H3 H4 I3 I4
(c) S-clusters
S1 S2
S3 S4S3 S4
S2S1
Dynamic Forwarding for 2D (1) Event may span multiple cells in a F-cluster
Assume the region spanned by an event is contiguous. Maximum 4 cells
(a) 1 Cell (a) 2 Cells (a) 3 Cells (a) 4 Cells
No other F-cluster will have packets Forward to sink
Forward to other S-aggregators
Dynamic Forwarding for 2D (2) Forwarding Rules
Rule 0: Forward packets to F-aggregator by structure-free data aggregation protocol.
Rule 1: Event spans three or four cells in a F-cluster, forwards to sink.
Rule 2: Event spans a cell in a F-cluster, forward to a S-aggregator.
F-cluster
Corresponding S-cluster
Cell generating packets
Dynamic Forwarding for 2D (2)
Rule 3: Event spans two cells, forward to two S-aggregators in order.
C1 C2
F-cluster X
F-cluster Y
S-cluster I S-cluster IIC C
Forward to 1st S-aggregator (near sink), then forward to 2nd S-aggregator
Sink
F-aggregator
S-aggregator
Aggregator Selections
Nodes play the role of F-aggregator in turn. With probability based on residual energy Hash current time to a node within that cluster
Delegate the role of S-aggregator to F-aggregator Select the F-aggregator in the F-cluster near sink as the S-aggregator
Sink
F-aggregator and
S-aggregator (Right-top S-cluster)
Sink
Dynamic Forwarding for 2D (3) Property 2. Packets will be aggregated at the F-
aggregator, at the 1st S-aggregator, or at the 2nd S-aggregator.
Experiments (1)
Experiments Environment 105 Mica2-based nodes 7 x 15 grid network Node spacing: 3 feet Transmission range: 2 grid-neighbor 2 F-clusters Fixed event location
Protocols Dynamic Forwarding over ToD (ToD) Data Aware Anycast (DAA) Shortest Path Tree (SPT) Shortest Path Tree with Fixed Delay (SPT-D)
Experiments (2)
Event Size
SourcesngContributiofNumber
onsTransmissiTotalofNumber
onTransmissi ofNumber Normalized
Better Performance:More chance of being aggregated
Long Stretch Problem
Experiments (4)
Large Simulation Environment 2000m x 1200m area 1938 nodes (grid network) Node spacing: 35m Transmission range: 50m Cell side length = Event diameter Event with random way-point model at 10m/s for 400 seconds
Protocols ToD DAA SPT OPT
Experiments (6)
Scalability (Event with different distance to sink)
Event Size: 400m Event Area: 400m x 800m Area Distance to Sink
: 200m ~ 1400m
Experiments (7)
Cell Size
Event Size:
200m, 400m, 600m
Best Cell Size:200m Event 100m Cell
400m Event 200m Cell
600m Event 200m Cell
Future Work:
Select appropriate cell size