Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst...

30
Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks 1

Transcript of Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst...

Page 1: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Aruna BalasubramanianBrian Neil Levine

Arun VenkataramaniUniversity of Massachusetts,

Amherst

Enhancing Interactive Web Applications in Hybrid Networks

1

Page 2: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

MotivationMobile users increasingly want to access network

applications on the go

Cost

Technologies

Application support

3G

Opportunistic WiFi

2

Page 3: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Our workHow can we support a broader class of interactive

applications using opportunistic WiFi access?

Internet

1. Mobile-to-Infrastructure contacts

2. Mobile-to-Mobile contacts

3

Page 4: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Application design space

Disconnection duration with AP

Conn

ectio

n du

ratio

n w

ith A

P

~1s ~60s ~3600s

~1s

~3600s

~60s

VoIP

Interactive Web?Web browsing,Web search

Delay TolerantEmail, bulk transfer m2m contacts

useful

m2i contacts useful

???

4

Page 5: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

OutlineHow can opportunistic mobile-to-Infrastructure

(m2i) contacts be used to support web search?

Can performance be improved by simultaneously leveraging mobile-to-mobile (m2m) contacts?

What is the performance of web search using opportunistic WiFi access?

5

Page 6: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Web search process<your favorite search engine>

Retrieving….

Retrieving….

6

Page 7: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Challenges: Intermittent connectivity

<your favorite search engine>

Retrieving….

7

Page 8: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Thedu: Adapting Web search for intermittently connected networks1. Use aggressive prefetching web pages to convert

interactive process to transactional

2. Prioritize prefetched web pages during bandwidth limited connection

8

Page 9: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Internet

1. Aggressive prefetching

TheduProxy

9

Page 10: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

1. Aggressive prefetching

Queries from mobile node

Store queryStore query Interface

Google, Yahoo, Live ,

Ask, ….

Google, Yahoo, Live ,

Ask, ….

SnippetsSnippets Prefetch Prioritized web pages

Prioritized web pages

Web pages returned to mobile node

Related work• Use prefetching to improve

availability[Coda91, Jiang98, Chandra01]

• Proxy-based to mask disconnections [Seth06,Ott06] 10

Page 11: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

2. Prioritizing useful web pages

• How many of these web pages are useful?• How many web pages to prefetch?

• How to compare web pages from different queries?

<your favorite search engine>

11

Page 12: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

How many to prefetch and how many web pages are useful? 80% users are interested in top 20 web pages

Thedu: Only prefetch webpages for top 20 url snippets

For some queries (homepage), there is only a single useful response

E.g., query “mobicom 2008” likely needs one response “http://www.sigmobile.org/mobicom/2008/”

Thedu: Identify homepage queries and send only one relevant web response

12

Page 13: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Homepage versus Non-homepage queries

Mobicom 2008<url> www.sigmobile.org/mobicom/2008 </url>

<snippet> MobiCom 2008, the 14th Annual International Conference on Mobile, </snippet>

<title> ACM Mobicom 2008 </title>

Mobicom 2008<url> www.sigmobile.org/mobicom/2008 </url>

<snippet> MobiCom 2008, the 14th Annual International Conference on Mobile, </snippet>

<title> ACM Mobicom 2008 </title>

Oil prices summer<url> www.eia.doe.gov/steo </url>

<snippet> Residential natural gas prices over the same period are projected to ,</snippet>

<title> EIA - Short-Term Energy Outlook </title>

Oil prices summer<url> www.eia.doe.gov/steo </url>

<snippet> Residential natural gas prices over the same period are projected to ,</snippet>

<title> EIA - Short-Term Energy Outlook </title>13

Page 14: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Thedu’s query-type classifier

Homepage Non Homepage

Query terms occur in URL Query is in question form

All query terms occur in title or snippet

Top URL is wikipedia

Less than 3 words Length greater than 3 words

URL is root

Thedu’s query-type classifier accuracy: 88%

14

Page 15: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

How to compare web pages from different queries?

Today’s search engine rank web page for a single query using relevance scoresScores not comparable across queries

Queries q1, q2q1

q2

Thedu’s query normalization technique

Thedu’s query normalization technique

Thedu: Aggressive prefetching1.Prefetch top 20 URLs for each query2. Identify home page queries and return only 1 relevant web page in expectation3.Prioritize the remaining web pages across queries by normalizing relevance scores

Thedu: Aggressive prefetching1.Prefetch top 20 URLs for each query2. Identify home page queries and return only 1 relevant web page in expectation3.Prioritize the remaining web pages across queries by normalizing relevance scores

15

Page 16: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

OutlineHow can intermittent mobile-to-Internet (m2i)

connectivity be used to support web search?

Can performance be improved by simultaneously leveraging mobile-to-mobile (m2m) contacts?

What is the performance of web search using opportunistic WiFi access?

16

Page 17: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Leveraging opportunistic m2m contacts

Internet

When useful?When meeting opportunities are skewed

Why useful? Because of decreasing marginal utility of web pages

17

Page 18: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Should nodes download their own web pages or route for others?

Thedu uses utility-based routing.

W

X

Y

y1y2…

x1x2…

w1w2…

w1w2… x1x2w3y1Sorted accordingto utility

18

Page 19: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Utility computation: IR meets networkingGoal: Maximize number of relevant responses delivered within a deadline

Utility of routing X or Y’s web pageP(web page is relevant) xP(W can deliver the web page to

destination within deadline)

Utility of downloading own web pageP(web page is relevant) xP(web page will NOT be delivered to W

within deadline if the opportunity is missed)

W

X

Y

19

Page 20: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

m2m routing between mobile nodes

Deliver web pages destined to peer

Route other web pages using similar utility-based routing

Exploit query locality by caching popular web pages

W

X

Thedu leverages m2m contacts 1. Using a utility-driven routing protocol 2. Exploiting caching

Thedu leverages m2m contacts 1. Using a utility-driven routing protocol 2. Exploiting caching

20

Page 21: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

OutlineCan intermittent mobile-to-Internet (m2i)

connectivity be used to support web search?

Can performance be improved by simultaneously leveraging mobile-to-mobile (m2m) contacts?

How does Thedu perform in practice?

21

Page 22: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Evaluation goalsDoes Thedu improve performance of web search

for opportunistic WiFi networks?

Is there a benefit for leveraging m2m contacts?

Evaluation based on deployment of Thedu on DieselNet testbed and trace-driven simulationsEvaluation based on deployment of Thedu on

DieselNet testbed and trace-driven simulations

22

Page 23: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Evaluation tools: Networking and IRDieselNet Testbed at UMass

Search engine10 GB web collection, Large user study

23

Page 24: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

DieselNet: Both m2i and m2m contacts

m2m meeting

m2i meeting24

Page 25: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Web search deployment results

RelevantWeb pagesdelivered

25

Page 26: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Delay in receiving relevant web page

26

Page 27: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Effect of m2i meeting frequency on web search performance

Deployment: March, 2007

Trace-driven: Nov, 2007

27

Page 28: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Leveraging m2m contactsOn DieselNet, leveraging m2m contacts provide

little benefitRatio of m2i versus m2m contacts is 20:1

Our analytical model confirms resultLeveraging m2m routing is useful only if AP density is

low

28

Page 29: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

Leveraging m2m contacts: Sparse AP

Leveraging m2m contactsprovides throughput benefit, but not delaydoes not improve performance of interactive applications,

even in sparse AP scenarios 29

Page 30: Aruna Balasubramanian Brian Neil Levine Arun Venkataramani University of Massachusetts, Amherst Enhancing Interactive Web Applications in Hybrid Networks.

ConclusionsThedu enables interactive web search application for

intermittently connected networks

Leveraging m2m contacts only provides throughput benefits and only when AP density is low

Deployment of Thedu on DieselNet shows 3 times improvement in number of relevant web pages

trace.cs.umass.edutrace.cs.umass.edu30