Reliable Multihop Transfer on Wireless Sensor Networks Rodrigo - - PowerPoint PPT Presentation

IEEE SECON – October 2004

Reliable Multihop Transfer

• n Wireless Sensor

Networks

Rodrigo Fonseca, Sukun Kim, David Culler

University of California Berkeley

IEEE SECON – October 2004

Motivation

Some sensornet applications

require 100% reliability over multiple hops

Structure monitoring Logging (development,

deployment)

Auditing

This has proven to be non-

trivial

IEEE SECON – October 2004

Challenges

Wireless communications

Low power radios Asymmetric, changing links Interference, etc.

Resource constrained hardware

Memory, computational power, energy

IEEE SECON – October 2004

Problem Scope

Design options for achieveing high reliability

• ver multiple hops

Traffic pattern:

One destination, large data (in comparison to pkt) Focus on convergence and point-to-point

Assumption:

Routing layer provides a path, or set of paths to

the destination

IEEE SECON – October 2004

Design Options

Only a fraction of the transmissions goes through

any given link

We can improve reliability by increasing

Number of packets injected Probability of success

Redundancy

Retransmissions -- End to End, Link Level, Both Erasure coding

Probability of Success

Path selection, Alternate paths Congestion Control

IEEE SECON – October 2004

Outline

Introduction Alternatives for Reliability

Retransmission Erasure Coding Alternate Routes

Experimental Results Conclusions Alternatives for Reliability

IEEE SECON – October 2004

End-to-End Retransmission

Probability of success over multiple hops

decreases rapidly

Wasted effort

Fail at hop n, n-1 wasted transmissions

E2E path may not exist at all times Reverse path may not exist

Although for 100% reliability source must receive

some signal from destination

IEEE SECON – October 2004

Link Level Retransmission

Found to be very efficient in increasing

reliability

Effect of boosting each link success probability Local repair

IEEE SECON – October 2004

Link Level Retransmission

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 Maximum number of retransmission Success Rate

Empirical Theoretical

Testbed experiment, 5 hops, avg link quality 77%

IEEE SECON – October 2004

Link Level Retransmission

Found to be very efficient in increasing

reliability

Effect of boosting each link success probability Local repair

However, still fails to reach 100% reliability

Bursty loss pattern

Cost of achieving even higher reliability may

become very high

IEEE SECON – October 2004

Erasure Codes

What if we can tolerate the loss of a few

percent of the packets?

Transmit redundant information Erasure codes allow M out of M+N packets to

be recovered

Fraction of redundancy is called rate of code

There are rateless codes, which can produce

unlimited redundancy, but may be expensive

IEEE SECON – October 2004

Encoding Channel Decoding M 8 msgs N 21 code words N’ =8 code words M 8 original msgs

Erasure Codes

IEEE SECON – October 2004

Benefit: if receiver has codes containing original messages

• Encoding, Decoding are faster
• Even if receiver get less than 8 packets, we don’t lose every

message

Systematic Codes

IEEE SECON – October 2004

Implementation on TinyOS

We use systematic codes No memory overhead for encoding or

decoding

Codewords generated on the fly Reception can stop once M pkts received

Real time operation on Mica2 motes Available for TinyOS

IEEE SECON – October 2004

But Losses are Bursty...

If we loose more than N-M packets, can’t

recover the entire data

Codes introduce a fixed redundancy

• verhead

So, depending on the loss process

Waste bandwidth on all packets Not effective when needed...

IEEE SECON – October 2004

Alternative Routes

Find Alternative Route: a form of ‘spatial retransmission’ But this may get tricky if we get a lot of failures Get k best candidates for the next hop from routing layer, and try from the best

IEEE SECON – October 2004

Alternative Routes

Dynamic alternative route selection

Provides immediate reaction to failed route

We change the routing layer to provide

possible next hops, instead of one

Successively try alternatives May still drop packet if no possible route

works

IEEE SECON – October 2004

Outline

Introduction Alternatives for Reliability

Retransmission Erasure Coding Alternate Routes

Experimental Results Conclusions Experimental Results

IEEE SECON – October 2004

Experimental Setup

Point-to-point routing

Beacon Vector Routing used to provide routes,

remained stable

Soda Hall Testbed, 78 motes 1 pair of nodes at a time, 300 packets @ 1/s

Results shown:

1 pair of nodes, 300 packets @ 1/s Average route 5 hops Other pairs similar results

IEEE SECON – October 2004

Testbed

Destination Source

IEEE SECON – October 2004

Metrics

We are mainly concerned with two metrics: Reliability

Fraction of application data packets that are received

by the destination

‘Work’

Number of transmissions per successfully received

packet, per hop

Ideally, 1 transmission per hop per message

IEEE SECON – October 2004

Reliability

0.2 0.4 0.6 0.8 1 1.2 1 2 3 4 5 5+AR Maximum number of retransmissions Success Rate

2 4 8

Erasure code redundancy Redundancy

IEEE SECON – October 2004

Work per packet, per hop

0.5 1 1.5 2 2.5 3 3.5 4 1 2 3 4 5 5+AR Maximum number of retransmissions Work (per packet, per hop) 1 2 4 8 AR means alternate route is used

Redundancy

IEEE SECON – October 2004

Reliability versus work

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5 3 3.5 4 Work (packets per received data packet, per hop) Reliability 1 2 5 5+AR

8 2 1 2 3 4 5 1 3 8

Retransmissions Redundancy

IEEE SECON – October 2004

Finding the best combination

Given a threshold reliability requirement, what is the retransmission /redundancy combination that has the smallest overhead?

Work

IEEE SECON – October 2004

Conclusions and Future Work

Main Contributions

We evaluated different combinations of options for multihop

reliability

Implementation of real time erasure coding on TinyOS

Combination of options yields best results

Erasure coding allows packet drops Alternate route makes the loss process more amenable to

erasure coding

Important for routing layer to quickly detect and route around

failure

Future work

Throrough characterization of loss patterns in other settings Experiments with different routing algorithms