improving on-demand data access efficiency with cooperative caching in manets phd dissertation...

33
Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense [email protected] Yu Du Chair: Dr. Sandeep Gupta Committ ee: Dr. Partha Dasgupta Dr. Arunabha Sen Dr. Guoliang Xue Supported in part by NSF grants ANI-0123980, ANI- 0196156, and ANI-0086020, and Consortium for Embedded Systems.

Upload: philip-may

Post on 14-Jan-2016

224 views

Category:

Documents


8 download

TRANSCRIPT

Page 1: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs

Phd Dissertation Defense

[email protected]

Yu Du

Chair: Dr. Sandeep Gupta

Committee: Dr. Partha Dasgupta

Dr. Arunabha Sen

Dr. Guoliang Xue

Supported in part by NSF grants ANI-0123980, ANI-0196156, and ANI-0086020, and Consortium for Embedded Systems.

Page 2: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

2

Roadmap

1. Introduction

2. Cooperative caching

3. Related works

4. Proposed approach – COOP

5. Performance evaluation

6. Conclusions and future works

Page 3: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

3Data Server

Data Server

1.1 Problems of data access in MANETs

• MANETs – Mobile Ad hoc Networks

– Wireless medium

– Multi-hop routes

– Dynamic topologies

– Resource constraints

• On-demand data access – client/server model.

Data Server

Data Server

Data Server

Data Server

Data Server

Data Server

1. Introduction

Page 4: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

4

1.2. Reducing data access costs in MANETs

• The locality principle [Denning]– Computer programs tend to repeat referencing a subset of

data/instructions.

– Used in processor caches, storage hierarchies, Web browsers, and search engines.

• Zipf’s law [Zipf]– P(i) 1/∝ iα(αclose to unity), common interests in popular data.

– 80-20 rule: 80% data accesses happen on 20% data.

• Cooperative caching– Multiple nodes share and cooperatively manage their cached contents.

1. Introduction

Page 5: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

5

1.3. Cooperative caching• Cooperative caching

– A caching node not only serves its own data requests but also the requests from others.

– A caching node not only stores data for its own needs but also for others.

– Shorter path, less expensive links, less conflictions, lower risks of route breakage.

– save time, energy, and bandwidth consumption as well as improves data availability.

• Why?– Data locality and commonality in users’ interests.

– Client/Server communication Vs. inter-cache communication.

– Users around the same location tend to have similar interests.

• People gathered around the food court: menus.

• Exploration team: environmental information.

1. Introduction

AshleyBob To remote server

Page 6: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

6

Roadmap1. Introduction2. Cooperative caching

2.1. Overview2.2. Cache resolution2.3. Cache management2.4. Cache consistency control

3. Related works4. Proposed approach – COOP5. Performance evaluation6. Conclusions and future works

Page 7: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

7

2.1 Overview

• Cooperative caching– Multiple nodes share and cooperatively manage their

cached contents.

– Cache resolution

– Cache management

– Cache consistency control

• Used in Webcache/Proxy servers on Internet.– To alleviate server overloading and response delay.

– Did not consider special features of MANETs.

2. Cooperative caching

Page 8: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

8

2.2 Cache resolution

• How to find a cache storing the requested data?

1

54

32

76?

?

OK

Hierarchical Directory-based Hash table based

Caching node Data items

Node 1

Node 2

Item1

Item2

1

2

3

4

5

Node 1

Node 2 Node 3

Harvest [Chank96]

Summary[Fan00]

Squirrel[Lyer02]

2. Cooperative caching

Page 9: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

9

2.3 Cache management

• What to cache?

– Admission control.

– Cache replacement algorithm.

• LRU

• Extended LRU (Squirrel)

– any access has same impact, whether it is from the local node or other nodes.

2. Cooperative caching

Page 10: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

10

2.4 Cache consistency control

• How to maintain the consistency between server and cache?

– Strong/Weak consistency: whether consistency is always guaranteed.

– Pull/Push-based: who (client/server) initiates the consistency verification.

• TTL is used in this research.

– Each data item has a Time-To-Live field – allowed caching time.

– TTL is popularly adopted in real applications – HTTP.

– Lower cost than strong-consistency protocols.

Pull-based Push-based

Weak TTL Synchronous Invalidation

Strong Lease Asynchronous Invalidation

2. Cooperative caching

Page 11: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

11

3. Related works

Schemes Cache Resolution Cachemanagement

Consistencycontrol

Networkmodel

Harvest [Chank96]

Hierarchically No specification TTL WAN

Summary[Fan00]

Directory-based LRU TTL WAN

Squirrel[Lyer02]

Hash-based Extended LRU TTL LAN

Cao04[Cao04]

CacheData,CachePath,

HybridCache

LRU TTL MANET

Page 12: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

12

Roadmap

1. Introduction

2. Cooperative caching

3. Related works

4. Proposed approach – COOP4.1. System architecture

4.2. Cache resolution

4.3. Cache management

5. Performance evaluation

6. Conclusions and future works

Page 13: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

13

4.1 System architecture

• Each node runs a COOP instance.

• The running COOP instance– Receives data requests from user’s

applications.

– Resolves requests using the cocktail cache resolution scheme.

– Decides what data to cache using COOP cache management scheme.

– Uses the underlying protocol stack. COOP

User Applications

Cocktail Cache Resolution

COOP Cache Management

User requests Response

COOP Messages

Network Protocol Stack

Local Cache

Secondarydata

Primarydata

Node A

4. Proposed approach – COOP

Page 14: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

14

4.2. Cache Resolution

4.2.1. Hop-by-Hop

4.2.2. Zone-based

4.2.3. Profile-based

4.2.4. COOP cache resolution – a cocktail approach

4. Proposed approach – COOP

Page 15: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

15

4.2.1 Hop-by-Hop cache resolution

• The forwarding nodes try to resolve a data request before relaying it to the next hop.

• Reduces the travel distance of requests/replies.

• Helps to avoid expensive/unreliable network channels.

Data Server

Data Server

Data Server

Data Server

4. Proposed approach – COOP, 4.2 Cache resolution

Page 16: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

16

4.2.2 Zone-based cache resolution

• Users around the same location tend to share common interests.

• Cooperation zone – the surrounding nodes within r-hop range.– r: the radius of the cooperation zone

• To find an item within the cooperation zone– Reactive approach – flooding within the cooperation zone.

– Proactive approach – record previous heard requests.

Data Server

Data Server

Data Server

Data Server

Data Server

Data Server

4. Proposed approach – COOP, 4.2 Cache resolution

Page 17: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

17

4.2.3 Profile-based cache resolution

• Records received request to assist future cache resolution– RRT – Recent Request Table.

– Entry is deleted when if the recorded requester fails to reply the corresponding data item.

– When the table is full, use LRU to decide replacement.

Requester Time Requested Data ID

192.168.0.11 15:26:59:08:16:2005 D1

192.168.0.15 15:25:59:08:16:2005 D2

192.168.0.18 15:20:59:08:16:2005 D3

4. Proposed approach – COOP, 4.2 Cache resolution

Page 18: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

18

4.2.4 COOP cache resolution – a cocktail approach

Received a datarequest

Data in localcache?

Success?

Reactive zone-basedcache resolution

Done

Yes

No

Profile-based cacheresolution

Hop-by-Hop cacheresolution

Success?Yes

Yes

No

No

4. Proposed approach – COOP, 4.2 Cache resolution

Page 19: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

19

4.3. Cache Management

4.3.1. Primary and secondary data

4.3.2. Inter-category and intra-category rules

4. Proposed approach – COOP

Page 20: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

20

4.3.1. Primary and secondary data

• Different cache misses may introduce different costs.– Example: cache miss cost for X is higher than cache miss cost for Y.

• Primary data and secondary data.– Primary data – not available within cooperation zone.

– Secondary data – available within cooperation zone.

Data Server

Data Server

X can be obtained from a

neighbor.

Y has to be obtained from

the server.

4. Proposed approach – COOP, 4.3 Cache management

Page 21: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

21

4.3.2. Inter-category and intra-category rules

• Inter-category rule– when replacement decision is to be made between different categories.

– Primary data have precedence over secondary data

• Intra-category rule– when replacement decision is to be made within the same category.

– LRU

• Example: A1 – A5 (Primary); B1 – B6 (Secondary)

T0 T4T3T2T1

4. Proposed approach – COOP, 4.3 Cache management

Page 22: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

22

Roadmap1. Introduction2. Cooperative caching3. Related works4. Proposed approach – COOP5. Performance evaluation

5.1. The impact of different zone radius5.2. The impact of data access pattern5.3. The impact of cache size5.4. Data availability5.5. Time cost: average travel distance5.6. Cache miss ratio5.7. Energy cost: message overhead

6. Conclusions and future works

Page 23: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

23

5.1 The impact of different zone radius

(1) Average probability of finding a requested item d within the zone.(2) Average time cost

– assuming time cost is proportional to the number of covered hops

(3) Average energy cost– assuming time cost is proportional to the number of messages.

2

)1(1)( rdPrP

2

)1())(1)(()( rdPLrrPLrrrPD

2

)1())(1)(()( 222 rdPLrrPLrrPrE

(1)

(2)

(3)

Pd average probability of a node caches d.

ρ the average node density.

L the distance (hops) between the requesting node and the server.

r the cooperation zone radius.

5. Performance evaluation

Page 24: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

24

5.1 The impact of different zone radius

2

)1())(1)(()( rdPLrrPLrrrPD

Average t i me cost Vs. cooperati on zone radi us

0

10

20

30

40

1 2 3 4 5 6 7 8 9 10

Cooperati on zone radi us

Aver

age

time

cos

t

Average energy cost Vs. cooperat i on zoneradi us

050

100150200250300350

1 2 3 4 5 6 7 8 9 10

Cooperat i on zone radi us

Aver

age

ener

gy c

ost

2

)1())(1)(()( 222 rdPLrrPLrrPrE

5. Performance evaluation

P( r) Vs. cooperat i on zone radi us

00. 20. 40. 60. 8

11. 2

1 2 3 4 5 6 7 8 9 10

Cooperat i on zone radi us

P(r)

2

)1(1)( rdPrP

• If an item is not found within a certain size cooperation zone, it is unlikely to find it within a larger size zone.• The saturation point.

Page 25: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

25

5.2 The impact of access pattern

Cache mi ss rat i o Vs Data access pat tern

0

0. 2

0. 4

0. 6

0. 8

1

1. 2

0. 5 0. 6 0. 7 0. 8 0. 9

Al pha

Cac

he m

iss

rati

o

SC

HBH

CT- 1

CT- 2

CT- 3

Average t ravel di stance Vs Al pha

0

2

4

6

8

10

0. 5 0. 6 0. 7 0. 8 0. 9

Al pha

Dist

ance

(Hop

s) SCHBHCT- 1CT- 2CT- 3

Average #messages Vs Al pha

0

5

10

15

20

25

0. 5 0. 6 0. 7 0. 8 0. 9

Al pha

#Mes

sage

sSCHBHCT- 1CT- 2CT- 3

• α ++• Cache miss ratio - -

•CT-3, CT-2, CT-1, HBH, SC• Average travel distance - -

•CT-3, CT-2, CT-1, HBH, SC• Average #messages - -

•HBH CT-1, SC CT-2, CT-3

5. Performance evaluation

Page 26: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

26

5.3 The impact of cache size

Average #messages Vs. Cache si ze

0

5

10

15

20

25

10 20 30 40 50

Cache si ze

Aver

age

#mes

sage

s

SCHBHCT- 1CT- 2CT- 3

Average t ravel di stance Vs Cache si ze

0

2

4

6

8

10

10 20 30 40 50

Cache si ze

Dist

ance

(hop

s) SCHBHCT- 1CT- 2CT- 3

Cache mi ss rat i o Vs Cache si ze

0

0. 2

0. 4

0. 6

0. 8

1

1. 2

10 20 30 40 50

Cache si ze

Cac

he m

iss

rati

o

SC

HBH

CT- 1

CT- 2

CT- 3

• Cache size ++• Cache miss ratio - -

•CT-3, CT-2, CT-1, HBH, SC• Average travel distance - -

•CT-3 CT-2, CT-1, HBH, SC• Average #messages - -

•HBH CT-1, SC CT-2, CT-3

5. Performance evaluation

Page 27: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

27

5.4 Data availability

Request success Rat i o Vs. node number

0

0. 2

0. 4

0. 6

0. 8

1

50 60 70 80 90 100

Number of Nodes

Succ

ess

rati

o

SCHBHCT- 1CT- 2

Request uccess rat i o Vs. node vel oci ty

0

0. 2

0. 4

0. 6

0. 8

1

1 10 20 30

Speed

Succ

ess

rati

oSCHBHCT- 1CT- 2

Request success rat i o Vs. pause t i me

0

0. 2

0. 4

0. 6

0. 8

1

60 120 180 240 300

Pause t i me

Succ

ess

rati

o

SCHBHCT- 1CT- 2

• Varied factors• node number• pause time• node velocity

• Data availability• CT-2, CT-1, HBH, SC

5. Performance evaluation

Page 28: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

28

5.5 Time cost: average travel distance

Average travel di stance Vs. node number

0

2

4

6

8

50 60 70 80 90 100Number of nodes

Numb

er o

f ho

ps

SCHBHCT-1CT-2

Average t ravel di stance Vs. pause t i me

0

24

6

8

60 120 180 240 300

Pause t i me

Numb

er o

f ho

ps

SCHBHCT- 1CT- 2

Average t ravel di stance Vs. nodevel oci ty

02468

1 10 20 30

Speed

Numb

er o

fho

ps SCHBHCT- 1CT- 2

• Varied factors• node number• pause time• node velocity

• Average travel distance• CT-2, CT-1, HBH, SC

5. Performance evaluation

Page 29: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

29

5.6 Cache miss ratio

Cache mi ss rat i o

00. 20. 40. 60. 8

11. 2

60 120 180 240 300

Pause t i me

Cach

e mi

ss r

atio

SCHBHCT- 1CT- 2

Cache mi ss rat i o Vs. node number

00. 20. 40. 60. 8

11. 2

50 60 70 80 90 100

Number of nodes

Cach

e mi

ss r

atio

SCHBHCT- 1CT- 2

Cache mi ss rat i o

00. 20. 40. 60. 8

11. 2

1 10 20 30

Speed

Cach

e mi

ss r

atio

SCHBHCT- 1CT- 2

• Varied factors• node number• pause time• node velocity

• Cache miss ratio• CT-2, CT-1, HBH, SC

5. Performance evaluation

Page 30: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

30

5.7 Energy cost: average #messages

Message overhead Vs. node number

05

1015202530

50 60 70 80 90 100

Number of nodes

Numb

er o

f me

ssag

es

SCHBHCT- 1CT- 2

Average number of messages per request

05

1015202530

60 120 180 240 300

Pause t i me

Numb

er o

f me

ssag

es

SCHBHCT- 1CT- 2

Average #messages per successf ul request

05

1015202530

1 10 20 30

Speed

Numb

er o

f me

ssag

esSCHBHCT- 1CT- 2

• Varied factors• node number• pause time• node velocity

• Average #messages• CT-1 HBH, SC, CT-2

5. Performance evaluation

Page 31: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

31

6. Conclusions and future works

• Cooperative caching is supported by data locality and the commonality in users’ interests.

• Proposed approach – COOP– Higher data availability – Less time cost– Smaller cache miss ratio – The tradeoff is message overhead– Tradeoff is dependent the cooperation zone radius.

• Future works– Adapt cooperation zone radius based on user’s requirements.– Explore different cooperation structure.– Enforce fairness in cooperative caching.

Page 32: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

32

References

• [Cao04] L. Yin and G. Cao, “Supporting cooperative caching in ad hoc networks”, INFOCOM, 2004.

• [Chank96] A. Chankhunthod et al. “A Hierarchical internet object cache”, USENIX Annual Technical Conference, 1996.

• [Denning] P. Denning, “The locality principle”, Communications of the ACM, July 2005.

• [Fan00] L. Fan et al. “Summary cache: A scalable wide-area web cache sharing protocol”, Sigcomm, 1998.

• [Lyer02] S. Lyer et al. “Squirrel: A decentralized peer-to-peer web cache”, PODC, 2002.

• [Zipf] G. Zipf, “Human behavior and the principle of least effort”, Addison-Wesley, 1949.

Page 33: Improving On-demand Data Access Efficiency with Cooperative Caching in MANETs Phd Dissertation Defense 11.21.05@CSE.ASU Yu Du Chair:Dr. Sandeep Gupta Committee:Dr

33

Q & A

Thank You!