30
UNIVERSITY OF NIVERSITY OF MASSACHUSETTS ASSACHUSETTS , A , AMHERST MHERST Department of Computer Science Department of Computer Science R3: Robust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics Xiaozheng Tie, Arun Venkataramani, Aruna Balasubramanian University of Massachusetts Amherst University of Washington

R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

  • Upload
    keon

  • View
    31

  • Download
    2

Embed Size (px)

DESCRIPTION

R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics. X iaozheng Tie, Arun Venkataramani, Aruna Balasubramanian U niversity of Massachusetts Amherst U niversity of Washington. W ireless routing compartmentalized. - PowerPoint PPT Presentation

Citation preview

Page 1: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS, A, AMHERST • MHERST • Department of Computer Science Department of Computer Science

R3: Robust Replication Routing in Wireless Networks with Diverse Connectivity

Characteristics

Xiaozheng Tie, Arun Venkataramani, Aruna Balasubramanian

University of Massachusetts AmherstUniversity of Washington

Page 2: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 2

Wireless routing compartmentalizedProtocols designed for well-connected meshes

OLSR, ETT, ETX, EDR, …

Protocols designed for intermittently-connected MANETs

AODV, DSDV, DSR, …

Protocols designed for sparsely-connected DTNs

DTLSR, RAPID, Prophet, Maxprop, EBR, Random, …

Research question: Can we design a simple routing protocol that ensures robust performance across networks with diverse connectivity characteristics all the way from well-connected meshes to mostly-disconnected DTNs and everything in between?

Page 3: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 3

OutlineCompartmentalized design harmfulQuantifying replication gainR3 design and implementationEvaluationConclusion

Page 4: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 4

Fragile performanceProtocols perform poorly outside target environment

DTN protocols perform poorly in mesh

Replication wasteful

Mesh protocols perform poorly in DTNs

No contemporaneous path

Mesh testbed DTN testbed

2.1x

Norm

alize

d de

lay

2.2x

Norm

alize

d de

lay

Page 5: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 5

Spatial connectivity diversityDieselNet-Hybrid

Vehicular DTN + Wifi Mesh20 buses in Vehicular DTN4 open AP WiFi mesh clusters

< 100 contacts

100 – 200 contacts

> 200 contacts

Page 6: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 6

Temporal connectivity diversityHaggle

Mobile ad hoc network8 mobile and 1 stationary imotes9 hour trace in Intel Cambridge Lab

Frac

tion

of

conn

ecte

d no

des

Page 7: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 7

Compartmentalized design harmfulFragile performance under spatio-temporal diversity

Makes interconnection of diverse networks difficult

ManageabilitySeparation of concernsLong-term innovation

Page 8: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 8

OutlineCompartmentalized design harmfulQuantifying replication gainR3 design and implementationEvaluationConclusion

Page 9: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 9

Replication: Key difference

DTN MeshMANET

Sparsely connected Well connectedIntermittently connected

Replication Forwarding

Key question: Under what conditions and by how much replication improves performance?

Page 10: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Model to quantify replication gain

10

Replication gain

μ =min E[X1],E[X2],...,E[Xn ]{ }

Src

X1

Dst

X2

Xn

μ(1) = E[min{X1,X2,...,Xn}]

• Delay of forwarding

• Delay of replication

μμ(1)

X i

Random variable denoting the delay of path i

Page 11: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 11

Example of replication gain

11

Replication gain

μ =min E[X1],E[X2]{ } =min 1,3{ } =1

Src

X1Dst

X2

μ(1) = E[min{X1,X2}] = 0.2

• Delay of forwarding

• Delay of replication

μμ(1)

= 5

P(X1 = 0.1) = 90%P(X1 =10) =10%E(X1) =1

P(X2 = 0.3) = 90%P(X2 = 30) =10%E(X2) = 3

Replication gain depends on path delay distributions, not just expected value

5x delay improvement

Page 12: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 12

Trace-driven analysis on DieselNet-DTN and Haggle

Replication gain vs. number of paths

Vehicular DTN in DieselNet Haggle

Two paths suffice to capture much of the gain

Page 13: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 13

Compartmentalized design harmfulQuantifying replication gainR3 design and implementationEvaluationConclusion

Outline

Page 14: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 14

R3 design overviewLink-state

Estimate per-link delay distribution

ReplicationSelect replication paths using modelAdapt replication to be load-aware

Source routing along selected path(s)

Y1

Y2

Y3

Src Dst

Page 15: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 15

Link delay

Estimate link delay using periodic probes

Estimate link delay distribution

Time

30s20s

10s0.1s

Link availability delay Delay to successfully transfer pkt

Delay samples = {30.1s, 20.1s, 10.1s}

Acked probe Half of round-trip delay

Unacked probe Half of time since sending probe and

receiving an ack for subsequent probe

Page 16: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 16

R3 design overview

Src Dst

Link-stateEstimate per-link delay distribution

ReplicationSelect replication paths using modelAdapt replication to be load-aware

Source routing along selected path(s)

Page 17: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 17

First pathPath s.t. it minimizes Selected using Dijkstra’s shortest path algorithm

Second pathPath s.t. it minimizesSelected using delay distributions and model

Path selection using model

k( ∈{1,2,...,n})

E[X k ]

E[min{X i,X k}]

k( ∈{1,2,...,n})

X1

X2

Xn€

X iSrc Dst

i

j

Page 18: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 18

ProblemReplication hurts performance under high load

SolutionLoad aware replication

Adapting replication to load

ForwardingReplicationStart

actual_delay > 2 * model_estimated_delay

actual_delay ≤ 2 * model_estimated_delay

Page 19: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 19

R3 design overviewLink-state

Estimate per-link delay distribution

ReplicationSelect replication paths using modelAdapt replication to be load-aware

Source routing along selected path(s)

Src Dst

Page 20: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 20

Compartmentalized design harmfulQuantifying replication gainR3 design and implementationEvaluation

Deployment on a DTN and mesh testbedSimulation based on real tracesEmulation using mesh testbed

Conclusion

Outline

Page 21: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 21

DieselNet DTN testbed20 buses in a 150 sq. mile area

Mesh testbed16 nodes in one floor

Simulator validation using DieselNet deployment

< 10% of deployment result

R3 Deployment

Page 22: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 22

Experimental settingsTemporal diversity inherent in HaggleSpatial diversity inherent in DieselNet-HybridVarying load

Compared protocolsReplication: RAPID, RandomForwarding: DTLSR, AODV, OLSRMulti-configuration: SWITCH (RAPID+OLSR)

R3 Trace-driven simulation

Page 23: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Robustness to spatial diversitySimulation based on DieselNet-Hybrid trace

23

R3 improves median delay by 2.1x

1

2

34

5

6

7

8

9

Page 24: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Robustness to varying loadSimulation based on DieselNet-Hybrid trace

24

R3 reduces delay by up to 2.2x over SWITCH

Page 25: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 25

Compartmentalized design harmful

R3 ensures robust performance across diverse connectivity characteristics

Unified link metric based on delay distributionsReplication based on delay uncertainty modelAdaptive replication based on network load

Conclusion

Page 26: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Q&A

26

Page 27: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 27

Theorem: Replication gain is high iff path delays are highly unpredictable

Predictability of a random variable X = Smallest such that

When is replication gain high?

ε

P[X ≤ εE[X]] ≥1−ε

Corollary: Replication can yield unbounded gain even with two paths

Page 28: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 28

Path delay Expected delay: Delay distribution of : convolutions of

Estimating path delay distribution

Y1

Yi

Y2

Yn

X

E[X] = E[Y1]+ E[Y2]+ ...+ E[Yn ]

X =Y1 +Y2 + ...+Yn

Y1,Y2,...,Yn

X

Link delay distribution Path delay distribution

Page 29: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Robustness to temporal diversity Simulation based on Haggle trace

29

R3 reduces delay by up to 60%

R3 increases goodput by up to 30%

Page 30: R3: R obust Replication Routing in Wireless Networks with Diverse Connectivity Characteristics

UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science

Emulating intermediate connectivity

Mesh-based emulation approach Brings link up and down

to vary connectivity Emulates connectivity

diversity (but not mobility)

30

R3 reduces delay by up to 2.2x