Title: Funneling-MAC: A Localized, Sink-Oriented MAC For Boosting Fidelity in Sensor Networks
1Funneling-MAC A Localized, Sink-Oriented MAC
For Boosting Fidelity in Sensor Networks
Gahng-Seop Ahn, Emiliano Miluzzo, Andrew T.
Campbell, Se Gi Hong, Francesca Cuomo Columbia
University, Dartmouth College, University La
Sapienza SenSys 2006 Presenter Tsung-Han Lin
2Outline
- Background
- Funneling effect
- Protocol design
- Evaluation
- Conclusion
3MAC in Sensor Nets
- Collision avoidance
- Increase throughput
- Duty cycling
- Energy efficiency
- Latency
4Contention based
- S-MAC (Infocom 2002)
- Coordinated radio on-off
- Energy-latency trade-off
- Looser time sync demand than TDMA
- T-MAC (SenSys 2003)
- Enable longer sleep period of S-MAC if no traffic
is detected
5Contention based
- B-MAC (SenSys 2004)
- Low power listening
- Flexible check period can be adjusted based on
app traffic pattern
6Contention based
- SCP-MAC (SenSys 2006)
- Reduce preamble with synchronous check period and
send time - Allow ultra-low duty cycle
- X-MAC (SenSys 2006)
- Shorter preamble
7TDMA based
- TRAMA (SenSys 2003)
- Distributed algorithm
- Maintains two-hop neighbors info
- High message overhead
- DESYNC (IPSN 2007)
- Self organized protocol
- Auto adjust slot size
- Very low complexity, high channel utilization
8TDMA/CSMA Hybrid
- Z-MAC (SenSys 2005)
- CSMA in low contention TDMA in high contention
- DRAND distributed coloring algorithm to assign
slots
9Funneling Effect
- The majority of packet loss occurs within the
first few hops from the sink
10Quantifying funneling effect
(1.5m)
- Dartmouth testbed 45 Mica2 nodes,16 random
sources - Delivery ratio 1 hop neighbor gt 80, 2 hop
neighbor lt 20 - MintRoute link quality based routing, data-ack
approach - B-MAC
11Traffic rate vs. Throughput
- To understand the impact under different network
load - 0.2 pps (light), 1 pps (medium), 4 pps(overload)
12Impact of funneling effect
- Overall loss rate 67 to 95
- Hop 1 and 2 have high loss rates
- 80- 90 of losses happen within the first 2 hops
from the sink! - Still true for light loaded network
13Any MAC resolves this?
- Z-MAC probably
- High contention TDMA
- However, Z-MAC schedules the whole network, which
is too costly - Funneling-MAC
- The scheduling is localized to sink
- Reacting dynamically to network condition
14The idea
- Hybrid TDMA/ CSMA scheme inside the intensity
region - Pure CSMA - pure CSMA scheme outside the
intensity region - Sink oriented TDMA scheduling
- Maintenance of the intensity region dynamically
operated by the sink
15Protocol Design
- On-demand beaconing
- Sink-oriented scheduling
- Dynamic-depth tuning
- Meta-schedule advertisement
16On demand beaconing
- To decide the size of intensity region
- To synchronize nodes in the region
17On demand beaconing
- Sink broadcast beacon periodically
18On demand beaconing
- Sink broadcast beacon periodically
- Nodes receive beacon as f-nodes, also synchronize
the clock
19On demand beaconing
- Sink broadcast beacon periodically
- Nodes receive beacon as f-nodes, also synchronize
the clock - Decide path head with data flow
Path heads
Path info is registered in the sink
20On demand beaconing
- Sink broadcast beacon periodically
- Nodes receive beacon as f-nodes, also synchronize
the clock - Decide path head with data flow
- If sink can schedule more f-nodes, increase the
transmission power
21On demand beaconing
- Sink broadcast beacon periodically
- Nodes receive beacon as f-nodes, also synchronize
the clock - Decide path head with data flow
- If sink can schedule more f-nodes, increase the
transmission power - Repeat (2)
Determined by dynamic depth tuning
22Sink-oriented scheduling
- Assign slots based on path and traffic
- Round robin for each path
- Sink broadcasts the TDMA schedule to every f-node
- Traffic measurement
- pkts / scheduled frame
- Moving average
- Spatial reuse
- The end of this path can share the same slot with
the head of next path if they are 3-hop away
23Sink-oriented scheduling
- Path A 4-hop
- Path B 4-hop
- Path C 3-hop
4
1
2
5
Spatial reuse Path B longer than 3-hop, 1 slot
can be reused
8
3
6
9
10
4
7
24Sink-oriented scheduling
4
1
2
5
8
3
6
9
10
4
7
25Sink-oriented scheduling
- If more traffic are presented
26Framing
- The schedule is carried with beacons
- CSMA control pkts, unregistered event data
- Synchronous LPL for f-nodes
27Dynamic-depth tuning
- Schedule more nodes, less loss?
- Probability of collision decrease with hops
- No
- Amount of TDMA slots are limited over scheduling
still leads to collisions - An optimal depth lies in between
28Dynamic-depth tuning
- Amax max number of slots that can be assigned
given the TDMA capacity - A number of slots required to schedule
path-heads traffic - if A lt Amax then sink increases beacon
transmission power - if A gt Amax then sink decreases beacon
transmission power
29Meta-schedule advertisement
- Reduce interference from possible interferer
- Nodes within the intensity region but do not
receive beacons due to radio reasons - Nodes close to the boundary of intensity region
- Recover if any intermediate nodes do not receive
beacon
30Meta-schedule advertisement
- Embed mini schedule in the data packet
- Interferer knows when the CSMA slot starts
- Beacon loss can recover from this
- Only 4-byte
- Superframe, TDMA frame, time left of current
frame, superframe repititions - Only in the first packet each beacon interval
31Evaluation
- Depth tuning
- Boundary node interference
- Loss rate distribution
- Multi-hop throughput
- Energy tax and signaling overhead
32Setup
- Compared with B-MAC and Z-MAC
- MintRoute routing
33Impact of depth tuning
- Traffic? optimal depth?
- Dynamic depth tuning is in need
34Impact of boundary node interference
- Fixed power -7dBm
- Variable power -6-8dBm
- 8 nodes in the boundary area
35Loss rate distribution
- Reduce the loss rate in the first two hops
36Throughput
- All 44 nodes as sources with 5 pps (heavy load)
- MintRoute take 20mins to establish paths
- If no path, the data would go broadcast
- Z-MACs performance degrades
37Schedule drift
- DRAND schedule may not be valid over time
- 76.7 nodes have change in neighbor tables.
- The change is around 25-45
- Performance fall back to B-MAC
- Periodic DRAND has large overheads
38Varying workload
- Schedule drift
- More slots to nodes closer to sink
MintRoute fail to setup routing path
39Energy tax
- The total amount of cost to deliver a bit per
node - Control overhead
- B-MAC no
- Z-MAC local time sync packet
- Funneling-MAC beacon pkt, schedule pkt, path
info header, meta-schedule header
40Signaling overhead
41Conclusion
- Mitigating funneling effect through scheduling in
the intensity region, even under lightly loaded
traffic. - Funneling-MAC outperforms Z-MAC and B-MAC in
various network conditions
42Reference
- Some slides from Emiliano Miluzzos talk in
SenSys 2006