The problem bit
This presentation is the property of its rightful owner.
Sponsored Links
1 / 19

The Problem Bit PowerPoint PPT Presentation


  • 120 Views
  • Uploaded on
  • Presentation posted in: General

The Problem Bit. Matthias Keller , Jan Beutel , Lothar Thiele. DCOSS 2013, Cambridge, MA, USA. Multi-hop Data Collection. Sensor nodes generate own packets and forward packets of other nodes. 1. 2. 3. Packet Information. FIFO Send Queue. Routing Tree. Source address

Download Presentation

The Problem Bit

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


The problem bit

The Problem Bit

Matthias Keller, Jan Beutel, Lothar Thiele

DCOSS 2013, Cambridge, MA, USA


Multi hop data collection

Multi-hop Data Collection

Sensor nodes generate own packets and forward packets of other nodes

1

2

3

Packet Information

FIFO Send Queue

Routing Tree

  • Source address

  • Sequencing information

  • (Imperfect) timing

  • information

  • Parent information

Optional

packet buffer

A

Error Model

4

Finite size

send queue

  • Lossy channel

  • Congestion

  • Backpressure

  • Protocol errors

Maintenance of system health and lifetime?


Extraction of internal network state

Extraction of Internal Network State

In-band

Extracted state is limited to

connected operation

Transmission

overhead

eScan

[Zhao2002]

Sympathy

[Ramanathan2005]

PAD

[Liu2011]

Hybrid

Monitoring

PDA

[Römer2009]

Passive/non-intrusive

Active/intrusive

FlockLab/GPIO[Lim2013]

LiveNet

[Chen2008]

Debugging/

development

MoteLab/printf

[Werner-Allen2005]

Out-of-band

… and many more


Packet delay de composition

Packet Delay (De)Composition

RX

RX

TX

TX

TX

penalty

2

2

1

1

2

Time

Time

End-to-end packet delay = Sojourn time at node H

= f(processing delay,

time without connection,

delay due to full parent,

# of transmission failures)


Multi hop network tomography

Multi-Hop Network Tomography

Passive method for the reconstruction of path and per-hop timing information:

F

F

A

C

A

C

Multi-hop Network Tomography

B

S

D

S

B

E

D

E

1

1

2

2

1

  • Incoming packets

  • Information frompacket

  • source

  • Order of arrival at the sink

  • Per-packet network path

  • Per-hop arrival order

  • Per-hop arrival time bounds

M. Keller, J. Beutel, L. Thiele: How Was Your Journey? Uncovering Routing Dynamics in

Deployed Sensor Networks with Multi-hop Network Tomography, SenSys 2012


Per hop sojourn time uncertainty

Per-hop Sojourn Time Uncertainty

Dozer, IPI = 120sec, 96 Tmote Sky nodes (TWIST)

Extra waiting time after

a failure event

Tomography data alone cannot

provide accuracy needed for

inferring failure events

N. Burri, P. von Rickenbach, R. Wattenhofer: Dozer: Ultra-Low Power Data Gathering in

Sensor Networks, IPSN 2007


Hybrid monitoring

Hybrid Monitoring

What is the minimal extra information needed to infer internal network state

with a high confidence?

  • Incoming packet stream

  • Timing information

  • Sequencing information

  • Parent information

Number of failure events that

occurred within a specific time

interval

Single extra bit per packet

Failure event: Event that causes waiting packets to wait longer.


Hybrid monitoring system

Hybrid Monitoring System

Hybrid Monitoring for the estimation of the number of failure events that occurred at a sensor node within a specific time interval:

Sensor Node:

Backend:

Problem Bit

Component

Multi-hop Network

Tomography

Failure Event

Count Estimation

Nodes add one extra bit to every

locally generated packet.

Minimum

Path Delay Estimation


Problem bit operation on a single node

Problem Bit Operation on A Single Node

Set problem bit

on error

Clear problem bit after

packet generation

1

Problem bit

0

FAIL

FAIL

0

1

OK

OK

OK

OK

OK

Time

RX

TX

GEN

TX

RX

TX

RX

TX

TX

TX

GEN

TX

Also waiting longer due to failed

transmission of previous packet


Hybrid monitoring system1

Hybrid Monitoring System

Hybrid Monitoring for the estimation of the number of failure events that occurred at a sensor node within a specific time interval:

Sensor Node:

Backend:

Problem Bit

Component

Multi-hop Network

Tomography

Failure Event

Count Estimation

Nodes add one extra bit to every

locally generated packet.

Minimum

Path Delay Estimation


Minimum path delay estimation

Minimum Path Delay Estimation

Idea: If a packet cannot have been delayed by a failure event, use its end-to-end

delay to improve the estimation of the minimum delay of its path.

For each node N that the packet under inspection k visited:

Could k have been delayed by a problem at N?

0

0

Time

Problem bit never set:

0

1

One or more problem bit set:

of interest

Exit loop

Yes

Update delay estimation

Visited all nodes?

No

No update


Hybrid monitoring system2

Hybrid Monitoring System

Hybrid Monitoring for the estimation of the number of failure events that occurred at a sensor node within a specific time interval:

Sensor Node:

Backend:

Problem Bit

Component

Multi-hop Network

Tomography

Failure Event

Count Estimation

Nodes add one extra bit to every

locally generated packet.

Minimum

Path Delay Estimation


Failure event count estimation

Failure Event Count Estimation

No

Set number of occurred

failure events to 0.

Is problem bit set?

Yes

Find packet that was generated before k

Analyze the timing of all packets that may be affected

by one or more problems within the time of interest

k

Previously generated packet

?

1

Time of interest

Time

Set number of occurred failure events

to MAX(1, estimated event count)


Testbed evaluation flocklab twist

Testbed Evaluation (FlockLab+TWIST)

  • Real implementation on top of TinyOS and Dozer

  • IPI = 120 sec, 30 sec Dozer slots (PermaSense)

  • Failure events that postpone the next packet transmission to the next communication slot:

    • Missed beacon

    • Busy radio

    • Missed packet acknowledgment

    • Full parent queue

  • Out-of-band extraction of ground truth information

N. Burri, P. von Rickenbach, R. Wattenhofer: Dozer: Ultra-Low Power Data Gathering in

Sensor Networks, IPSN 2007


Comparison with completely passive system

Comparison with Completely Passive System

Completely passive system

Completely passive system over

and underestimates the number of

occurred failure events.

96 Tmote Sky nodes (TWIST), 17 hours


Correlation with ground truth

Correlation with Ground Truth

96 Tmote Sky nodes, 17 hours

Significant improvement

with problem bit

Low correlation without

problem bit


Hybrid monitoring1

Hybrid Monitoring

  • Event detection accuracy increases significantly after adding one bit of information to every packet

  • Hybrid monitoring requires only a negligible overhead while also covering disconnected operation

  • Amount of health status automatically scales up with more traffic

www.permasense.ch


How about 2 3 n bits

How About 2, 3, …, N Bits?

Use cases for more than one extra bit:

  • Separate error causes

    • e.g., extra bits for busy radio, missed beacon etc.

  • Improve temporal accuracy

    • e.g., dedicate one bit to every 30 sec long Dozer slot

  • Move towards counter variables

  • Highest gain is achieved by using one extra bit instead of a completely passive system

1

2

3


  • Login