BCP Slides

5-10-2010

Challenge the future

DelftUniversity ofTechnology

RoutingRouting Without Routes:The Backpressure Collection Protocol

Steffan Karger

Moeller et al.

2Routing without Routes: The Backpressure Collection Protocol

Presentation outline

• Introduction

• Backpressure routing

• Adjustments for BCP (Backpressure Collection Protocol)

• The weight function

• Queue optimization

• Implementation

• Results

• Conclusion


Introduction

• Network assumptions

• Sensor nodes

• Data sinks

• Collection Tree Protocol

• Beaconing

• Builds MST’s


Backpressure routingRouting without routes

Data packets flow to the sink in a natural fashion

Monticello Dam in northern California



Data packets flow to the sink in a natural fashion

• Link weight

• Transmission threshold



• Use ETX as link usage penalty:

• Minimizes ETX -> fast delivery




Adjustments for BCPFIFO vs LIFO (1)

• Under FIFO, the average source-to-sink delay grows with decreased loading, why?


Adjustments for BCPFIFO vs LIFO (2)

• Under LIFO, data packets cascade towards sink

• Average source-to-sink delay does not grow under low load


Adjustments for BCP

• Problem: Queue size grows with distance to the sink

• Large networks -> large queues

• Solution: introduce virtual queue

• Discard oldest packet in real queue

• Record virtual queue size

• Null packets

Floating queues


Adjustments for BCP

• With fixed queue size algorithm

breaks down

• Floating queue preserves

gradient

Floating queues


Implementation

• 23 KB footprint (vs 27 KB for CTP)

• Weight recalculation parameter τ = 50ms• Performance/processor load tradeoff

• Link metric estimation (ETXi->j and Ri->j )• Based on data transmissions and ACK’s

• Advertize queue size in data header• Neighborhood snoops headers


BCP vs CTP

BCP CTP

Forwarding policy Gradients fromneighborhood

ETX from neighborhood

Setup cost Initial packetloss -

Perturbation cost - Rebuilding MST

Gradient advertizing Header snooping Beaconing


Results

• Tutornet• 40 nodes, indoor• IEEE 802.15.4-based Tmote Sky devices• -18 dBm transmit power• Poisson traffic

• Static network tests

• External interference

• Highly mobile sinks

Experiment details


Results

• BCP performs comparable to CTP on delay and delivery order

Static network – Delay performance

Source to sink delay CDF at 0.25 PPS for motes 4 and 40

The Reordering Distance for BCP under FIFOand LIFO servicing priorities.


Results

• BCP performs better then CTP with higher loads

Static network – Goodput

Goodput against load


Results

• 802.11 interference• Alternating: 20s on, 20s off

• CTP: 55% - 84% delivery ratio• BCP: 88% - 96% delivery ratio

External interference


Results

• Sink ‘moves’ every second

Highly mobile sinks


Results

• Sink ‘moves’ every second

• BCP adjust very good and even profits from mobile sinks

Highly mobile sinks

MobilityBCP CTP

StaticBCP CTP

Delivery ratio 0.996 0.590 0.969 0.999

Average Tx/packet

1.73 9.5 2.39 2.65


ResultsHighly mobile sinks

A 200 second window of sink time versus source mote for sinks 8, 18 and 26

• Data is ‘dumped’ when sink

is nearby


Conclusion

• In static networks:

• BCP performs comparable to CTP on delay and efficiency

• BCP can cope with hihger loads

• Robust to external interference

• Very good performance with highly mobile sinks


Discussion

• What happens when we use a duty cycle operation?

• BCP should work quite well with synchronized duty-cycle MAC’s like S-MAC

• A-synchronous duty-cycle MAC’s like LPL MAC need adjustments for packet snooping

Duty cycled operation


Discussion

• How about transmission count?

• Comparable to CTP

Transmission count

Per source average per packet transmission count to the sink

Documents

BCP Slides