Learning with Purpose

SmartParcel: A Collaborative Data Sharing Framework for Mobile

Operating Systems

Bhanu Kaushik∗ Honggang Zhang†

Xinwen Fu∗ Benyuan Liu∗ Jie Wang∗

∗Department of Computer Science, University of Massachusetts, Lowell, MA.

†Department of Computer and Information Science, Fordham University, Bronx NY

Learning with Purpose

IntroductionMotivation and Related WorkProblem DefinitionArchitectureSimulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

IntroductionMotivation and Related WorkProblem DefinitionArchitectureSimulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

Huge number of Mobile Devices such as Smartphones, Tablets, PDAs, portable media players etc.“About 6.2 billion users around the globe” – Ericsson, 2012.These devices support large number of Internet based applications. These Applications work on simple one-to-one client-server data distribution model.

Introduction

Results in: • Increasing concerns about volume of global

online digital content generated by these devices.• Multi-fold increase in Network traffic originating

from these devices• “100 PetaByte/Month in 2007 to 700

PetaByte/Month in 2012”-Ericsson, 2012. • Huge incumbent content availability and

maintainability cost.

Learning with Purpose

Introduction

Motivation and Related WorkProblem DefinitionArchitectureSimulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

Major challenges faced by mobile Internet users• Carrier enforced limited data plans,• Unavailability of hardware (3G or LTE),• Unavailability of access points, • Service outages and• Network and server overloads.

Results in: • Unavailability of application data to the users• High service maintainability cost, to both the

service providers and hosting servers.

Motivation and Related WorkMotivation

Learning with Purpose

Proposed Solutions for data offloading • Large Scale

• Alvarion, “Mobile data offloading for 3G and LTE networks.” • Cisco, “Architecture for mobile data offload over Wi-Fi access networks.”

• Small Scale• Han et. al. “Mobile data offloading through opportunistic communications

and social participation”• Lee et. al., “Mobile data offloading: How much can wifi deliver?”

Unaddressed Issues:• Entail huge changes in both, state of the art software and

hardware technologies • Do not take into account the heterogeneity of application data.

Motivation and Related WorkRelated Work : Data Offloading

Learning with Purpose

Delay-Tolerant Networks (DTN)• Target the interoperability between and among

challenged networks Familiar Strangers • Coined by Stanley Milgram in 1972,

“Individuals that regularly observe and exhibit some common patterns in their daily activities”.

SmartParcel uses the idea of opportunistic connectivity and in-network storage and retransmission from DTN architecture to ensure data delivery among the nodes in a “Familiar Strangers” network set up.

Motivation and Related WorkRelated Work: Opportunistic Data delivery and Familiar

Strangers

Learning with Purpose

IntroductionMotivation and Related Work

Problem DefinitionArchitectureSimulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

Our Goal is to develop framework of a Mobile data offloading and Service Assurance scheme by encouraging collaborative data sharing among spatio-temporally co-existing mobile devices.

Problem DefinitionSmartParcel

Fig. 1 : Proposed SmartParcel Approach

Learning with Purpose

IntroductionMotivation and Related WorkProblem Definition

ArchitectureSimulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

Service Discovery ManagerData Transfer ManagerService Cache Manager• Dynamic Cache• Static Cache

Network Interface ManagerService APIsCentral Control Manager

ArchitectureComponents

Fig. 2 : SmartParcel Service Architecture.

Learning with Purpose

Service Discovery Manager:• Identifies the available candidates for data transfer by broadcasting a “SYN”

message periodically • “SYN” packet contains meta-data about applications registered to

SmartParcel.• The meta-data is organized as a key value pair, i.e., (“ApplicationId,

TimeStamp”). • At receiver, based on the meta-data information it sets up a one-to-one

connection

Data Transfer Manager: • Manages the data transfer.• Can manage concurrent connections to multiple devices.• To reduce the network overhead, sends data for multiple applications as one

chunk.

ArchitectureComponent Details

Learning with Purpose

Service Cache Manager: • Service cache to store the application specific (heterogeneous) data .• Dynamic Cache

• In-memory cache for storing the applications meta-data information. • Implemented as Hash Map with (Application Id, Timestamp) as key-value

pairs. • Static Cache

• Static cache for storing the actual application specific data. • Maintained as SQLite database. • Schema “Application Id (as string), Data (as blob), Time Stamp”• Primary key : Application id and timestamp• Flexibility to developer to assign “Time to live” and “Reset-Time” for the

application data, end of day by default.

ArchitectureComponent Details

Learning with Purpose

Central Control Manager: • Manage the control from all components of the SmartParcel service. • All components work under same instance for synchronous operation.

Network Interface Manager: • Internal service, responsible for managing network connections.• Assists Service Discovery for identifying available devices on different

network interfaces (3G, LTE, WiFi, BlueTooth etc.). Service APIs: • Subscribe or unsubscribe to service• Update app data• Settings• Sharing statistics etc.

ArchitectureComponent Details

Learning with Purpose

Android SDK• New set of permissions.• SMP_ALL, SMP_BLUETOOTH, SMP_WIFI,

SMP_NFC and SMP_BT_WIFI.

ArchitectureAndroid and SmartParcel

Group Name BlueTooth WiFi NFC DISk-IO

SMP_ALL √ √ √ √SMP_BLUETOOTH √ × × √

SMP_WIFI × √ × √SMP_NFC × × √ √

SMP_BT_WIFI √ √ × √Table 1 : Resources used in different permissions

Fig. 3 : Integration of SmartParcel in Android framework

Android OS• Integrated in the “System Server” module.• System Server is launched by Zygote.• Zygote forks the SmartParcel service as a system

service. • Ensures system level privileges and independence

from the application “context”.

Learning with Purpose

IntroductionMotivation and Related WorkProblem DefinitionArchitecture

Simulation SetupResults Conclusions and Future Work

Outline

Learning with Purpose

MIT Reality Mining Data Set• 100 unique devices, 500,000 hours, 9

months • We use the Bluetooth encounters data.

Simulation Setup

Encounters Activity

Maximum 65 901

Minimum 2 4

Mean 4 243

Std. Dev. 8.67 133 Fig 4 : Hourly Variation of Device Encounters.

Fig 5 : Distribution of Device Encounters. Fig 6 : Distribution of Active Devices Per Day.

Table 2 : Data Set Description

Data Set

Learning with Purpose

Data Refresh Rate (DRR) : The frequency with which the data is being refreshed. Allowed Server Connections (ASC) : Number of devices allowed to get data from server on each day. User Participation Probability (UPP) : The Probability of user acting selfish, i.e., limiting its participation by only receiving data and not sending data

We measure the Data Availability Ratio (DAR)

Simulation SetupSetup Parameters

Learning with Purpose

IntroductionMotivation and Related WorkProblem DefinitionArchitectureSimulation Setup

Results Conclusions and Future Work

Outline

Learning with Purpose

User Participation Probability (UPP) = 100%Data Refresh Rate (DRR) =1 Refresh interval

ResultsEffect of user’s social activity level

Fig 6 : Effect of ASC on DAR over the Day, when ASC = 1

Fig 7 : Effect of ASC on DAR over the Day, when ASC = 30

Fig 8 : Effect of ASC on DAR , when ASC =1 to 75 devices.

Learning with Purpose

User Participation Probability (UPP) = 100%Data Refresh Rate (DRR) = 2 Refresh intervals, 12:00am -11:59am and 12:00pm-11:59pm

ResultsEffect of Data Refresh Rate (DRR)

Fig 9 : Variation of Data Availability Ratio (DAR) with Data RefreshRate (DRR) when DRR = 2 and Refresh Interval 12:00 am - 11:59 am.

Fig 10 : Variation of Data Availability Ratio (DAR) with Data RefreshRate (DRR) when DRR = 2 and Refresh Interval 12:00pm - 11:59pm.

Learning with Purpose

User Participation Probability (UPP) = 100%Data Refresh Rate (DRR) = 3 Refresh intervals.

ResultsEffect of Data Refresh Rate (DRR)

Fig 11 : Refresh Interval 12:00am-07:59am.

Fig 12 : Refresh Interval 08:00am-03:59pm.

Fig 13 : Refresh Interval 04:00pm-11:59pm.

Learning with Purpose

User Participation Probability (UPP) = 10%, 20%, 50% and 90%.Data Refresh Rate (DRR) = 1 Refresh intervalAllowed Server Connections(ASC) = 1 to 90 devices.

ResultsEffect of Selfishness

Fig 14 : Variation of Data Availability Ratio with User ParticipationProbability(UPP) and Allowed Server Connections(ASC).

(*Median of 1000 Simulation runs)

Learning with Purpose

IntroductionMotivation and Related WorkProblem DefinitionArchitectureSimulation Setup

Results Conclusions and Future Work

Outline

Learning with Purpose

“SmartParcel” - A novel approach for Data sharing among co-existing and co-located devices is presented.“One for all”, multiple incentive system for application developers, Internet service providers and application data providers (eg. cloud services) with collateral benefits for the consumer itself.We discussed the Design and implementation “SmartParcel” in Android. Implementation in android framework dictates the feasibility of the architecture.Flexibility of design ensures integration in almost every existing mobile operating system. In the future, we intend to investigate the scalability and performance issues encountered on real devices.

Conclusions and Future Work

Learning with PurposeLearning with Purpose

Thank You !Questions ?