71
1 Resource sharing in mobile wireless networks Maria Papadopouli Computer Science Department Columbia University http://www.cs.columbia.edu/~maria

1 Resource sharing in mobile wireless networks Maria Papadopouli Computer Science Department Columbia University maria

  • View
    213

  • Download
    0

Embed Size (px)

Citation preview

1

Resource sharing in mobile wireless networks

Maria PapadopouliComputer Science Department

Columbia Universityhttp://www.cs.columbia.edu/~maria

2

Academic background

• Columbia University Ph.D. candidate Fall 1996- advisor Prof. Golubchik Fall 1996–

1998 advisor Prof. Schulzrinne Fall 1998-• New York University M.S. Computer Science May 1994• University of Crete B.S. Computer Science June 1992

3

References on resource sharing

in mobile ad hoc networks

1. “Effects of power conservation, wireless coverage & cooperation on

data dissemination among wireless devices “, ACM MobiHoc 2001

2. “Performance analysis of 7DS a data dissemination & prefetching

tool for mobile users”, IEEE Sarnoff 2001, best paper/poster award

3. “7DS in mobile ad hoc networks”, Globecom 2000

4. “Performance of data dissemination among mobile devices”, journal submission, 2002

5. “Design & implementation of a P2P data dissemination &

prefetching tool for mobile users”, Metro 2001 6. “Network connection sharing in ad hoc wireless network among

collaborative hosts”, Nossdav 1999

with Prof. Schulzrinne

4

References on video on demand

7. "A Scalable Video on Demand server for a Dynamic Heterogeneous Environment", Lecture Notes in Computer Science, Springer 1998

8. "Support  of VBR Video Streams Under Disk Bandwidth Limitations", ACM SIGMETRICS Performance Evaluation Review 1997

9.  (with also J.C-S. Lui), "A survey of approaches to fault tolerant design of video on demand servers: Techniques, analysis and comparison", Special issue of Parallel Computing Journal on Parallel Data Servers and Applications 1998

with Prof. Golubchik

5

Outline

• Introduction– Background on wireless data access– Motivation– Overview of 7DS

• Performance analysis on 7DS• Conclusions • Future work

6

Background

• Fast growth in pervasive computing devices• Fast wireless data services growth• Base stations for wireless WAN will not keep

pace– Regulatory, environmental & cost barriers

for a dense deployment

Users experience intermittent connectivity & limited data access

7

Mobile information access

Dependency on infrastructure :

• Wireless WAN eg 802.11, 3G, CDPD, GSM, Bluetooth,

Ricochet• Infostations (Rutgers)

– When a client is in the proximity of the server, it access the data

• Peer-to-Peer– Routing in mobile, ad hoc & sensor networks

8

Mobile information access

Interactivity model :• Synchronous

– Users directly access or request the data

• Asynchronous (using prefetching)– Hoarding (Coda [CMU], Seer [UCLA])

9

Limitations of infostations & wireless WAN

•No communication infrastructure eg field operation missions, tunnels, subway

•Emergency•Overloaded •Expensive•Wireless WAN access with low bit

rates & high delays

10

Limitations of ad hoc networks

• All hosts cooperative• Complete path for the communication of two hosts

Host A Host B

11

Limitations of hoarding

• Only files• Files exist prior to disconnection • No dynamic generated information

12

Wireless data services

• Delay tolerant• Location-dependent services • User location hints at data needs• Overhead to discover, access &

update local data

13

Challenge

Accelerate data availability & enhance dissemination & discovery of information under bandwidth changes & intermittent connectivity to the Internet due to host mobility

considering power, bandwidth & memory constraints of hosts

14

Our Approach

Increase data availability by enabling devices to share resources

– Information sharing–Message relaying–Bandwidth sharing

• Self-organizing• No infrastructure• Exploit host mobility

15

Outline• Introduction

– Background on wireless data access – Motivation– Overview of 7DS

• Simulations & Analysis on 7DS– Information dissemination– Message relaying– Bandwidth sharing

• Conclusions • Future work

16

7DS

• Application• Zero infrastructure• Relay, search, share & disseminate information• Generalization of infostation • Sporadically Internet connected• Coexists with other data access methods• Communicates with peers via a wireless LAN• Power/energy constrained mobile nodes

17

Examples of services using 7DS

schedule info

WANWAN

autonomous cache

newsevents in campus,pictures

where is the closest Internet café ?

service location queries

traffic, weather, maps, routes, gas station

pictures, measurements

18

Information sharing with 7DS

Host B

Host C

data cache hit

cache miss

data

Host A

query

WANWAN Host A Host D

query

WLAN

WLAN

19

7DS options

Forwarding

Host A Host B

query

FWquery

Host C

time

Querying

active (periodic)

passive

Power conservation

on

off time

communication enabled

CooperationServer to client

Peer to peer server to client only server shares datano cooperation among clients• fixed info server (infostation model)• mobile info server peer to peer data sharing among peers

20

Outline• Introduction• Simulations & Analysis on 7DS

– Information dissemination– Message relaying– Bandwidth sharing

• wireless LAN• video on demand environment

• Conclusions • Future work

21

Simulation environment

pause time 50 smobile user speed 0 .. 1.5 m/shost density 5 .. 25 hosts/km2

wireless coverage 230 m (H), 115 m (M), 57.5 m (L)

ns-2 with CMU mobility, wireless extension & randway model

dataholder

querier

randway model

wireless coverage

22

Simulation environment

pause time 50 smobile user speed 0 .. 1.5 m/shost density 5 .. 25 hosts/km2

wireless coverage 230 m (H), 115 m (M), 57.5 m (L)

ns-2 with CMU mobility, wireless extension

pause1m/s

mobile host data holder

querierwireless coverage

23

Simulation environment

pause time 50 smobile user speed 0 .. 1.5 m/shost density 5 .. 25 hosts/km2

wireless coverage 230 m (H), 115 m (M), 57.5 m (L)

ns-2 with CMU mobility, wireless extension

v1

v2

v3

wireless coverage

data

24

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15 20 25

Density of hosts (#hosts/km )

Da

tah

old

ers

(%

) P2P data sharing(power cons.)

P2P data sharing

P2P data sharing & FW(power cons.)

Fixed Info Server

Mobile Info Server

Dataholders (%) after 25 min

high transmission power

2

Fixed Info Server

Mobile Info Server

P2P

25

Scaling properties of data dissemination

2 km

2 k

m

1 k

m

1 km

If cooperative host density & transmission power are fixed, data dissemination remains the same

R

R

wireless coverag

e

26

Scaling properties of data dissemination (cont’d)

R

R/2

For fixed wireless coverage, the larger the density of cooperative hosts, the more efficient the data dissemination

wireless coverage

27

Average delay (s) vs. dataholders (%)

0

200

400

600

800

1000

1200

0 5 10 15 20 25 30 35Dataholders (%)

Avera

ge D

ela

y (

s)

Fixed Info Server(medium transmission power) 4 initial dataholders (servers) in 2x2

Fixed Info Server (high transmission power ) one initial dataholder (server) in 2x2

one server in 2x2high transmission power

4 servers in 2x2medium transmission power

Fixed Info Server

28

Average Delay (s) vs Dataholders (%)Peer-to-Peer schemes

0200400600800

1000120014001600

0 10 20 30 40 50 60 70 80 90 100Dataholders (%)

Ave

rag

e D

elay

(s)

P2P (high transmission power) one initial dataholder & 20 cooperative hosts in 2x2

P2P(medium transmission power) one initial dataholder & 20 coperative hosts in 1x1

medium transmission power

high transmission power

29

Scaling properties of data dissemination

(cont’d)

Lr

R

Lwireless

coverage of info server

v xxr/2

R/2

v xx

30

Modeling Fixed Info Server as diffusion-controlled process

• trapping model with particles C and T (traps)

• particles C perform random walk in 2D space

• particles T static, randomly distributed in space of infinite capacity

• particles T absorb C when C step onto them

survival probability n at long times n

log (n) -An

querier particle C

fixed info server trap

trapping receiving dataC

T

31

Fixed Info Serversimulation and analytical

results

0

20

40

60

0 500 1000 1500 2000 2500 3000Time (s)

Dat

aho

lder

s (%

)

simulation model

Probability a host will acquire data by time t follows 1-e-at

high transmission power

32

Outline• Introduction

– Background on wireless data access– Motivation– Overview of 7DS

– Performance analysis on 7DS– Information dissemination– Message relaying– Network connection sharing

• Conclusions • Future work

33

Message relaying with 7DS

Host B

Messagerelaying

Host A

messages

Gateway

WAN

Host AWLAN

WLAN

34

Message relaying

• Take advantage of host mobility to increase throughput

• Hosts buffer messages & forward them to a gateway

• Hosts forward their own messages to cooperative relay hosts– Restrict number of times hosts forwards

35

0

20

40

60

80

100

5 10 15 20 25Density of hosts (#hosts/km )

Mes

sage

rel

ayed

(%

)

High transmission power (No FW)High transmission power (FW 6)Medium transmission power (No FW)Medium transmission power (FW 6)

2

Messages (%) relayed after 25 min (average number of buffered

messages : 5)

36

Outline• Introduction

– Background – Motivation– Overview of the system

• Performance analysis– Information dissemination– Message relaying– Network connection sharing

• Conclusions • Future work

37

Network connection sharing

WANWAN

Wireless LAN

Host A

Host B Host C

Host D

Hosts A & B dual-homedThey act as gateways to WAN for hosts C & D

Host E

Host F

thin WAN links

38

Network connection sharingprotocol

WANWAN

Host A

Host B Host C

Host D

Host E

1. C sends request for gateway

2. B & A respond advertising their bandwidth in WAN link

4. C selects least loaded gateway (eg A)

5. A C admission controlWLAN

thin wirelessWAN links

39

Benefits using network connection

sharing

• Statistical multiplexing for bursty traffic• Increase bandwidth utilization of the WAN links

– 80% bandwidth utilization for Pareto traffic– Load balancing across gateways

• For shared data applications :– Reduction of replicated data– Increase quality of service

40

Outline• Introduction

– Background on wireless data access – Motivation– Overview of the system

• Performance analysis– Information dissemination– Message relaying– Network connection sharing

• Conclusions • Future work

41

Conclusions Dominant parameters:

• density of cooperative hosts• wireless coverage density of cooperative

hosts & their mobility For fixed cooperative hosts density & transmission

power : scale area performance same For fixed wireless coverage density :

Density of cooperative host performance

42

Conclusions (cont’d) Probability a host will acquire data by time t in

• Fixed Info Server : 1-e-at

• Peer-to-Peer : 1-e-at

Message relaying is beneficial :

• Probability a message will reach the Internet • Utilization of available throughput by taking advantage of host mobility

43

Future work• Location-dependent applications & services• Actual traces & models for user mobility,

access patterns & data locality• Enhanced power conservation mechanism• Security & micro-payment issues• Extension of network connection protocol• Generalization of diffusion models for P2P• Adaptive scalable algorithms for information

discovery

44

Summary of contributions in video on demand

Novel multimedia retrieval scheduling algorithms

In multi-disk environments :• adapt to bandwidth changes • maximize data retrieval for all streams

using replication and multi-resolution

In single-disk environments :• allocate disk bandwidth in a fair manner

45

Thank you!

46

Future work: short term

• More on power conservation for data dissemination

• Peer-to-peer scheme using diffusion controlled processes

• Prototype– Deployment of 7DS in CU campus & in Bremen– Public release of the code

• Collaborations – IBM, HP, Bertelsmann & Limewire (Gnutella)

47

Future work : longer term

• Information discovery & dissemination in pervasive computing

– Model & abstractions for the quality of information

– Tight energy, bandwidth– Privacy & security for mobile, peer-to-

peer applications– Scaling & structural properties

48

Preventing DoS attacks

receives query

multicast query

Host Q Host R

multicast challenge

sends response

run non-trivialcomputationaltask

wait to hear ifQ is challenged

verifies Q’s answer

decides to cooperate

49

Electronic check payment

receive e-checkverify it is genuinestore e-check

Host Q Host R

send data

send e-check

wait for data from R

verify R is known to the bank &authorized for 7ds

send credentials

50

Token-based payment

receive query

Host Q Host R

send data

verify R’s public key

wait for data from R

check token countersend public key with report

form querysend query

decrease countersend ack increase token counter

decrease countersend nack increase token counter

send data

51

Information discovery & dissemination in pervasive

computing• Query & data locality

No need of infrastructure — use 7DS

• Query routing required Use infrastructure of gateways that create peer-to-

peer overlay hierarchies in self-organizing manner based on query demand & resources

[Castro, Greenstein, Muntz, Bisdikian, Kermani,

Papadopouli “Locating Application Data Across Service Discovery Domains”, MOBICOM’01]

52

7DS Implementation

• Cache manager (3k lines)• GUI server (2k lines)• HTTP client & methods (24k lines)• Proxy server (1k lines)• UDP multicast & unicast (1k)• Web client & server (2k)• Jar files used (xerces, xml,lucene, html

parcer)

53

020406080

100

5 10 15 20 25Density of hosts (#hosts/km )

Mess

age r

ela

yed (

%)

High transmission power (No FW)High transmission power (FW 6)Medium transmission power (No FW)Medium transmission power (FW 6)

2

Message relayed to gateway after 25 min

54

Network connection sharing summary

1) Requests for network connection

3) Gateway selection Load balancing

criteria

2) Advertisement of gateway availability

4) Admission control using Measured sum [Jamin et al]

u v+r v: measured load r: (peak) rate requested u:utilization target :bandwidth of WAN link

GatewayClient

55

Gateway selection mechanism

• Load balancing criteriaReduction of the maximum difference inthe average load over an interval across the

gateways :maxi{Li()}-mini{Li()}/Li(): average traffic measured at gateway

i over interval

• Greedy algorithm: Choose the least loaded gateway

56

Network connection sharing

BandwidthUtilization (%)

Pkt dropping rate (%)

Load balancingcriteria (%)

Exponential

66 0.002 2

Pareto 81 9 2

Pareto & exponential: 312 s(ON), 325s (OFF)Pareto, shape par. : 1.2 Flows: 64kb/s, 0.6 s int., avg hold time 5 minPerfect load balancing: 0%

57

Pareto traffic measurement policy

T(s), S(s)

Link Utilization(%)

Pkt lossRate (%)

60, 400 31 0.09

30, 400 37 0.2

3, 400

81 10

Larger T higher measured load more conservative admission

58

Information discovery & dissemination in pervasive

computing• Without infrastructure :

– 7DS exploits query & data object locality & host mobility

– Cooperation among hosts based on resources• With infrastructure :

– Gateways create peer to peer overlay hierarchies in self-organizing manner

– Participate based on query demand & resources

Castro,Greenstein,Muntz (UCLA), Bisdikian,Kermani(IBM), Papadopouli(Columbia Un.), “Locating Application Data Across Service Discovery Domains”, MOBICOM’01

59

Information discovery in pervasive

computing

• Dynamic nature of the environment: uncertainty, errors, timeliness & redundancy

• Local autonomy– Partial knowledge, local decisions to achieve a

global effect• Self-organization to minimize administration

overhead• Adaptive, scalable algorithms & protocols

Castro, Greenstein, Muntz (UCLA), Bisdikian, Kermani (IBM), Papadopouli (Columbia Un.),

“ Locating Application Data Across Service Discovery Domains”, MOBICOM 2001.

60

Epidemic model

• Carrier is “infected”, hosts are “susceptible”

• Transmit to any give host with probability ha+o(h) in interval h

• Pure birth process• T=time until data has spread among all

mobiles

• E[T]=1/a i=1

N-1

i(N-1)1

61

7DS implementation

• Initial Java implementation on laptop

• Compaq Ipaq (Linux or WinCE)• Inhand Electronics ARM RISC board

– Low power– PCMCIA slot for storage,

network or GPS

62

Mobility models

User mobility :• Randway • Random direction• Boundless simulation

area

• Gauss-Markov with history of previous

move

Group mobility• Column mobility• Pursue mobility• Nomadic community

mobility

63

Subway model

• Passengers arrive at subway stations– Poisson process 1/1-3min– ride : 2-6 stops– 1 min to leave the platform

• Subway line– 10 stops– Train with 6 cars– Arrives at a stop every 5 minutes

• Percentage of dataholders after they leave the subway for 1/ = 3 min is 65%

64

Types of attacks in ad hoc networks

Basic mechanisms• MAC layer• Routing mechanisms

– Malicious users agree to forward messages but fail to do so

– False routing information messages

– Selfishness & service enforcement issues

Security mechanisms• Distributed trusted server

under the control of malicious party

• Public key maliciously replaced

65

Service enforcement

• Lock out mechanism for selfish or misbehaving users– Denial of service attacks– Locked out node moves away where his

behavior is not reported

• Virtual micro currency mechanism– Incentives to cooperate– Discouraged from overloading the network

terminodes.org (EPFL), mojonation.net

66

Virtual micro currency

• Nodes remunerate each other for the services they provide to each other

terminodes.org (EPFL), mojonation.net

67

Information discovery & dissemination in pervasive

computing

• Dynamic nature of the environmentUncertainty, errors, timeliness & redundancy

• Local autonomyPartial knowledge, local decisions to achieve a

global effect

• Self-organization Minimize administration overhead

• Adaptive, scalable algorithms & protocols

68

Wireless WAN access

Location

what cost

UK 3G $590/person

Germany

3G $558/person

Italy 3G $200/person

New York

Verizon(20MHz)

$220/customer

• Spectrum is very expensive

• 3G bandwidth is very low (64kbs)

69

Avantgo: wireless service provider

70

Vindigo: wireless service provider

71

NYC wireless public infrastructure