Network Computing Laboratory MoB: A Mobile Bazaar for Wide-area Wireless Services R. Chakravorty, S. Agarwal, S. Banerjee, I. Pratt Mobicomm 2005

Network Computing Laboratory

MoB: A Mobile Bazaar for Wide-area Wireless Services

R. Chakravorty, S. Agarwal, S. Banerjee, I. Pratt

Mobicomm 2005

Network Computing Laboratory | 2

Korea Advanced Institute of Science and Technology

Roadmap

One line summary Introduction

Background Mobile Bazaar Service scenario

Basic concepts of MoB Applicable environment Services Pricing and reputation

MoB Architecture Implementation Evaluation

FTP application Web browsing application Location determination application Vito evaluation

Conclusions



One line summary

To overcome an existing impasse in mobile applications and services, this paper presents Mobile Bazaar (MoB), an open market, collaborative architecture to improve data services for wide-area wireless users.



Introduction: Background

Lack of wireless connectivity WLAN

Coverage is limited to specific spots Cellular data networks

Larger coverage However, poor connectivity in some areas

Coarse-grained competition Contract with only one provider Relatively Long-terms contract for all wireless data services.



Introduction: Mobile Bazaar (MoB) Defines fine-grained competition

Let users to choose and change among several providers at arbitrarily small timescales Choose the best provider for his immediate requirement Allow the user choose multiple providers at a time.

Open market structure Resell user’s unused resource as a service

Trader, consumer Any device can autonomously advertise services independent of all other devices

Incentive-based collaboration: monetary payoff Reputation and trust management (e.g. e-Bay) Supports composition of fine-grained service interactions

So? Integration of heterogeneous mobile terminals Better performance through Wireless Diversity Decouples infrastructure providers from services providers Customization for Diverse Applications



Introduction: Service Scenarios

Bandwidth aggregation Location Determination Web proxy caching

Time synchronization P2P data search



Basic concepts of MoB: applicable environments

Best scenarios is where there are many users in a small range: coffee-shop, shopping, malls, bus, train.

Coffee-shop: • 2/3 of the users spent more than 2 minutes in it• ~50% spent 10 minutes or more



Basic concepts of MoB: service & pricing and reputation

Service Location Protocol (SLP) Multi-hop interactions

Composed of multiple single-hop interactions (pair-wise) Simplify responsibility and pricing problem

Service composition Application layer

No need to change the network protocol behavior

No end-to-end path Consist of two independent TCP connections Different from previous add-hoc network work

Pricing and reputation Pricing is regulated by individual users

laissez faire approach

Reputation and trust management and billing system are required Vito (modeled on eBay)



MoB Architecture

3 Components Internet infrastructure

Providing internet connectivity Possible to be a trader in the MoB architecture

Mobile devices Formation of the dynamic network Service trading decision by user preference and policies

Third-party services for billing, reputation and trust management Vito Optional third-party services can be deployed

Bandwidth aggregation service

3 Mode of Operations incentive based with trust management incentive with trust assumption altruistic



Vito

Centralized system Like eBay system Hosted by third-party

Data forwarding scenario, where A wants to transfer at 30Kbps

K+ : public key K- : private key [m]K : encryption

message using key K TS : timestamp

Service request

SLP multicast address(239.255.255.253) Port 427 with TTL 1



Vito

Security mechanism Using public key cryptography Performance

Customer : 2 private key encryption, 1 public key verification Trader : 1 private key encryption, 2 public key verification Modulus length of 512 bits 40 ms and 2.5 ms for signature generation and verification on Pentium 2 266MHz machine

running Linux Reputation

Trader uploads its own positive feedback upload Natural result

Trader uploads positive feedback for customer Strong motivation for positive feedback

Customer uploads negative feedback for the trader No recourse for malicious customer, but eBay works anyway. Users are selfish, but not malicious

Customer pay prior to receiving service Trader has no proof of transaction for a danger of a malicious customer However customer has negative reputation feedback

Reputation fee No one can build up reputation for free



Implementation

Middleware installed in each device

Stripe manager Block-based application-level data

striping Dynamically changes the block size,

the number of parallel TCP connections, and connection types

Neighbor discovery layer Periodic scanning using link-specific

mechanism 802.11

One specific channel ofr neighbor discovery

Bluetooth interface Initiates scan procedure



Evaluating MoB application

File-transfer Indoors, devices equipped with Bluetooth, 802.11 cards, 3G cellular

interfaces :EvDO (2.4Mbps,153Kbps) and 1xRTT (144Kbps,64Kbps) The customer upload the file in blocks Churn

High churn : 10~20sec Med churn : 40~60sec Low churn : 60~120sec

Web browsing applications Collaborative location determination applications



Ftp scenario 1

Start with T1 T2 come to vicinity area at t1 Connection established with T2 at

t2 T1 is move away at t3



At most two traders Parallel download with single

bluetooth interface No effective gain in download

performance Eliminates ‘dead-time’

Only one trader at a time 5MB audio MP3, Using bluetooth

between devices Reason of block download

Payment Efficient data download with multiple

trader

Ftp Scenario 2,3: about churn



FTP scenario 4

At most two traders Parallel download with one bluetooth and one 802.11a

interface Significant performance benefits




File-transfer Web browsing applications

Using collaborative caching Static web contents caching

Collaborative location determination applications



Web Browsing for collaborative caching

Two MoB traders 1xRTT and EvDO

sites objects KB

Washingtonpost.com 61 250

Sourcefourge.org 41 122

Wisc.edu 46 67

Slashdot.org 26 129

E-auction 52 153




File-transfer Web browsing applications Collaborative location determination applications

Obtain location information without GPS devices Environment

4km x 3.2 km, 50 blocks in Manhattan, 1000 ~ 5000 cars Each vehicle moves at constant speed toward its destination

Different fractions of cars are equipped with GPS Communication range of trader and consumer

80 meter, 802.11b radio range with maximum transmission power of 30mW

Using other device’s (w/o GPS) location information



Collaborative location determination

With 5000 cars around, if only 5% of cars have GPS the accuracy is 72m Within one block

No intelligent technique Possibility of improvement Using speed of a vehicle’s

motion Using ‘anchors’



Evaluation of Vito

Internet access model Continuous connectivity Uniformly at random End of day connectivity

Selection policies based on Reputation Price Price with reputation threshold Ratio of reputation to price

Experiment Price reduction factor ν , Price increment factor λ, Price p no sell : p = p – νp, Sell : p = p + λp, ν = λ = 0.01 Requests up to 50 services



Evaluation of Vito



Evaluation of Vito



Evaluation of Vito



Evaluation of Vito



Conclusions

Although all originate idea have existed, really interesting and fascinating idea

Details on all aspects Detailed and rich scenarios and evaluations Business and legal concerns Security issues

Users would like it However, ISP’s wouldn’t like it

Still doubtful about performance Reputation check for every block requesting

Too Simple reputation & billing system It will not be deployed in Korea

Legal problem



Backup slides



Vito

Further design consideration for vito Sybil attacks – 이름 바꿔 들어오기

Entering fee, newbie 에게 불이익 Once-in-a-lifetime pseudonym

Collusion 서로 밀어주기 The abuser 어뷰저 in starcraft NP-hard problem, heristic approach P2P scenario, pre-trusted peers

Decentralized reputation management P2P scenario, pre-trusted peers

Documents

Network Computing Laboratory MoB: A Mobile Bazaar for Wide-area Wireless Services R. Chakravorty, S. Agarwal, S. Banerjee, I. Pratt Mobicomm 2005