Dissertation Proposal Aruna Balasubramanian Department of Computer Science, University of Massachusetts, Amherst Architecting Protocols To Enable Mobile.

Dissertation Proposal

Aruna Balasubramanian

Department of Computer Science, University of Massachusetts, AmherstArchitecting Protocols To Enable Mobile Application in Wireless Networks

Vision: Universal network accessMostly connectedIntermittently connectedMostly disconnected*Traditional wireless network protocols are not well-suited for disruption-prone environments

Thesis goalUnderstanding the principles underlying the design and implementation of a robust protocol stack to enable mobile applications in heterogeneous network environments.*

Research questionsWhat are the challenges in the diverse network environments taken in isolation?

How can we design decentralized mechanisms and algorithms to overcome challenges in each environment?

How can the algorithms adapt as a user moves between different environments?*

Diverse network environmentsEmail, bulk transferWeb search, Web browsingVoIP*

Evaluation methodologyDeploymentDieseNet, VanLAN

Trace-driven simulationsTraces collected from testbeds

AnalysisCompetitive analysis, Linear Programming and Modeling

*

RoadmapFundamental challenge: Tolerating disruption under uncertainty

*Mechanism: Opportunistic resource usageReplicationAggressive PrefetchingOpportunistic forwarding

Decentralized algorithm: Utility-driven prioritizationPacket prioritizationInformation prioritizationSender prioritization

Proposed work: Adapt as a user moves to different network environments or has access to multiple networks

DisruptionsMobile wireless networks are disruption proneCoverage holesMobilityChannel fading

Disruption results in a unique challenge in each environment

Tolerating disruptions challenging because of uncertain network conditions

*

Mostly disconnected networksNo connectivity to infrastructure

Uncertain network conditions

ii*

Intermittently connected networksFrequent disruptions due to coverage holes

*Internet

Mostly connected networksPacket level disruptions occur even when the mobile node is in range of an AP

*

RoadmapFundamental challenge: Tolerating disruption under uncertainty

*Mechanism: Opportunistic resource usageReplicationAggressive PrefetchingOpportunistic forwarding

Decentralized algorithm: Utility-driven prioritizationPacket prioritizationInformation prioritizationSender prioritization

Proposed work: Adapt as a user moves to different network environments or has access to multiple networks

RAPID: Replication to improve performance of disruption tolerant applications [Sigcomm 07]Mostly connectedIntermittently connectedMostly disconnected*

Why replication?Find paths under uncertainty*iiiii

How to replicate under resource constraints?Existing replication-based DTN routing protocolsUse a maximum replication countReplicate to nodes with better delivery probability

Metrics desired in practiceMinimize average delayMaximize packets meeting their deadlines

Incidental RoutingEffect of mechanism on routing metric unclear

Idea: Translate desired metrics to per packet utilities and replicate packets to intentionally improve utility*

UtilitiesUtility: expected contribution of packet to the metric. For example Minimize average delay, U(i) = negative expected delay of i

What is the improvement in utility by replicating a packetIf expected meeting times are exponentially distributed with mean Expected delay of a packet is Replicating the packet will reduce delay to /2

Use these utilities to design RAPIDResource Allocation Protocol for Intentional DTN Routing

*

RAPID Protocol: Utility driven prioritizationRAPID Protocol (X,Y):

1. Control channel: Exchange metadata

2. Direct Delivery: Deliver packets destined to each other

3. Replication: Replicate by prioritizing in decreasing order of marginal utility

4. Termination: Until all packets replicated or nodes out of range

Change in utilityPacket size*

A case for a heuristic solutionProblem: Given a set of transfer opportunities and packet workload, what allocation of packets will minimize delay?

With complete knowledge of workload and transfer opportunitiesSolving the DTN routing problem is NP Hard

With only knowledge of workload or transfer opportunitiesAn online DTN routing problem can be arbitrarily far from an offline adversary

*

Deployment on DieselNetDeployed RAPID on DieselNet for 58 days and validated simulator results with deployment results*

Trace-driven simulation results*

RoadmapFundamental challenge: Tolerating disruption under uncertainty

*Mechanism: Opportunistic resource usageReplicationAggressive PrefetchingOpportunistic forwarding

Decentralized algorithm: Utility-driven prioritizationPacket prioritizationInformation prioritizationSender prioritization

Proposed work: Adapt as a user moves to different network environments or has access to multiple networks

Thedu: Prefetching to improve web search performance [Mobicom 08, Chants 07]Mostly connectedIntermittently connectedMostly disconnected*

Web search challenge*

Retrieving.

Retrieving.

Prefetching and utility-driven prioritizationInternetPrefetch web pagesDeveloped two information retrieval techniques to prioritize web pages so that the most useful web pages are downloadedWebqueriesWebresponsesGoogle, Yahoo, Live ,Ask, .*

Deployment and resultsDeployed Thedu on DieselNet

Results from one week deployment The number of useful web pages delivered to the user using Thedu is 4 times greater than an existing protocol*

RoadmapFundamental challenge: Tolerating disruption under uncertainty

*Mechanism: Opportunistic resource usageReplicationAggressive PrefetchingOpportunistic forwarding

Decentralized algorithm: Utility-driven prioritizationPacket prioritizationInformation prioritizationSender prioritization

Proposed work: Adapt as a user moves to different network environments or has access to multiple networks

ViFi: Opportunistic forwarding to improve performance of interactive applicationsMostly connectedIntermittently connectedMostly disconnected*

Opportunistic forwarding

Tolerate packet-level disruptions

Internet

Problem: Coordinating among APs to forward the packet

*

Probabilistic coordinationGuidelinesRedundant relays should be minimizedThe intended next hop should receive the packet with high probabilityShould avoid per-packet coordination

Solution:APs relay an overheard packet probabilistically, such that the guidelines are satisfied*

ViFi deployment and resultsDeployed ViFi on VanLAN BSes and vehiclesCurrently 2 vans, 11 Bses2 months deployment

Deployment resultsViFi doubled the duration of a VoIP call compared to 802.11 by improving packet reception

RoadmapFundamental challenge: Tolerating disruption under uncertainty

*Mechanism: Opportunistic resource usageReplicationAggressive PrefetchingOpportunistic forwarding

Decentralized algorithm: Utility-driven prioritizationPacket prioritizationInformation prioritizationSender prioritization

Proposed work: Adapt as a user moves to different network environments or has access to multiple networks

Proposed workAdapting to different networks

Leveraging simultaneous access to multiple networks

Designing a self adapting protocol stack

*

Adapting to changing network environmentReplicationAggressive prefetchingOpportunistic forwarding*

Adapt replication to diverse network environments*Replicationimproves performanceDoes replication improve performance?

When is replication useful?Hypothesis: Replication is useful when (1) delay estimates have high variance and (2) metric is delay

*

Experimentally validating hypothesisMesh testbed at UMass increase uncertainty by removing/delaying links

DTN DieselNet testbed decrease uncertainty using bus schedules

*

How should packets be replicated?RAPID cannot be used as isIn our DTN model, transfer opportunities occur one after the otherIn other environments, node can have simultaneous transfer opportunities

Adapting RAPID for mostly connected networksWhich node to replicate to? How to take interference into account?Can we exploit opportunistic forwarding?

*

Proposed workAdapting to diverse networks

Leveraging simultaneous access to multiple networks

Self adapting protocol stack

*

Simultaneous access

If a node has both 3G and WiFi access3G networks have limited capacityOpportunistic WiFi access can be used to augment 3G capacity

*Research problemEmpirically quantify the connectivity and throughput of a WiFi network, to determine the extent of augmentation

Why quantify the connectivity and throughput?

Opportunity: WiFi can be used, to send background traffic, such as prefetched web responsesto aggregate available bandwidth to send datato send forward error correction bits and reduce losses

Challenge: WiFi networks are disruption prone and unpredictable

Approach: Quantifying will help determine how to use WiFi access

*

Empirical quantificationGiven a network, what is the average connectivity and throughput that can be supported?Connectivity Fraction of times the source can connect to the destinationThroughput Total traffic that can be sent between a source and destination

Challenge: Connectivity and throughput of a network depends on the network protocolsExample: A routing protocol can provide close to optimal connectivity in one network, but perform poorly in another

*

Performance of routing on different tracesThree network traces DieselNet, Muni San Francisco bus network, ETH traffic simulatorThree routing policies: Optimal, AODV, and GPSR*

Proposed methodology: Measurement and Analysis

Understand how network characteristics affect protocolsWhat network characteristics affect performance of a routing policy? How does the wireless range affect connectivity of a network?

How should opportunistic WiFi be used to augment 3G networks?

*

Proposed workAdapting to diverse networks

Leveraging simultaneous access to multiple networks

Self adapting protocol stack

*

What are the protocol layers?Traditional wireless stack not well-suited for disruption-prone networks*

Interface design802.11, Opportunistic forwardingRoutingLinkPre-fetchingApplication*Research questionsWhat information is exchanged between layers?What environment estimates are needed for the protocol layers to be self adapting?

End-to-End, Hop-by-HopForwardingContemporaneous, Replication DTN 2 TCP

TimelineMeasurement, Design, Implementation and Evaluation

Adapting replication 5 months

Leveraging simultaneous access 5 months

Self adapting protocol stack 2 months

Writing 2 months

*

ConclusionsThe vision is to provide universal network access

Proposed three algorithms to tolerate disruptions and enable applications RAPID -- mostly disconnectedThedu -- intermittently connected ViFi -- mostly connected networks

Proposed a protocol stack that adapts to changing network conditions*

Using backpressure algorithms for routing in uncertain networksNo contemporaneous end-to-end path

Send packets to a peer, if the peer has a smaller queue to the destination

Algorithm is throughput optimal

Can backpressure be adapted for minimizing delays?

*

How does wireless range affect connectivity?Should we use smaller number of long hops, or larger number of short hops

Short hopsHigher throughputLower SNR

Long hopsRouting overheadPath instability

*

Availability of WiFi vs 3G on DieselNet75% of grids that have 3G connectivity, also has WiFi connectivity*

Why WiFi?Urban areasCheaper alternativeCan augment capacity of existing technologiesRural areasEasy and cheap to deployThe deployment can grow organicallyOperates in the unlicensed spectrum, higher peak bandwidth, does not need expensive infrastructure*Todays WiFi protocols are not well suited to enable applications for mobile users

Completed workUnderstand challenges + Design mechanisms and algorithms =Sigcomm 08Sigcomm 07Chants 07, Mobicom 08*Routing protocol fordelay tolerant applicationsApplication layer protocolfor web applicationsLink layer protocol for interactiveapplications

Related work: PrefetchingPrefetching has been used To improve availability in file systems [Coda91, Chandra01]To improve performance of the web in wired networks, when the Internet was slow [Google Accelerator, Jiang98]In an email-based web browser [TEK]

Resource allocation was not the challenge in these environments

*

Related work: Opportunistic forwardingOpportunistic protocols for WiFi mesh (ExOR, MORE)Uses batching: Not suitable for interactive applications

Path diversity protocols for enterprise WLANs (Divert)Assumes APs are connected through a high speed back plane

Soft handoff protocols for cellular (CDMA-based)Packet scheduling at fine time scalesSignals can be combined*

Application development on DTNs to allow routing that can optimize a routing metric desired in practiceIncidental routing effects application development;Applications deployed on DTNs may desire some metrics to be optimized;How to adapt DTN oruitng mechanisms to different application needsProtocol where the routing policy can be tuned to suit the needs of an application;DO NOT USE ETC?Problem with existing routing protoco: Pause; Effect

RAPID borrows insights from existing DTN routing mechanisms;

RAPID draws upon insights from exisitng routing protocols; but we put them together in a way so as to enable the intentional routing propertyHeuristic that locally maximize the utility of replication per unit resource used