Reflections on ad-hoc and partially disconnected networks

Henning SchulzrinneSuman SrinivasanArezu Moghadam

Andy YuenColumbia University

Introduction

• Are ad-hoc and sensor networks the next active networks?

• What are the uses and users?

• What are missing pieces in the wireless puzzle?

Ad-hoc/sensor networks• More research interest than application interest:

– limited, mostly military applications– always repeat the same handful of examples

• vineyards, glaciers, …

– number of papers >> number of users• cf. active networks• brittle for regular users:

– easily splits into disconnected sub-networks– difficult to plan

• mesh networks: early experiences dubious– business model? (Singapore)– reliability and availability– frequency management in dense deployments

What’s missing?• Lots of practical problems• 802.11 configuration and debugging

– IETF experience: 1500 engineers can’t keep networks up and running

– manual channel assignment, no load balancing, gratuitous channel dropping application crash, long association delays

– no location information (cf. Skyhook) – security mechanisms

• something other than typing in 16 hex digits• opportunistic security and association (e.g., get token)

• 3G (IMS)– configuration– system complexity– new applications?

A set of predictions• WiMax for rural areas (water tower)• 3G/4G (= 3G without the PSTN legacy) in (sub)urban

areas and on major transportation corridors– easier to deploy than mesh– better power management– but hard to deploy for non-carriers

• 2.5G in rural areas• 802.11g/n indoors and as last-hop access

– cheap– on every laptop– reasonably fast– easy to deploy

Motivation

802.11 currently hard to deploy across city or large area

Problem: How can mobile devices / gadgets get information?

Peer-to-Peer data sharing Network

Solution: 7DS!

Wireless networksaccess (802.11)

cellular (3G)

mesh ad-hoc sensor 7DS

speed 10 Mb/s 1 Mb/s 500 kb/s 500 kb/s? 100 kb/s 10 Mb/s

ubiquity islands (100’)

urban urban islands (500’)

islands

(500’)

dispersed

density high high locally high

low

dataflow sink sink mesh mesh sink sinks

power high medium high low --

medium

low low

IllustrationIn the absence of the Internet, nodes can exchange information amongst themselves

Internet

7DS Overview

Information Dissemination and Resource Sharing

Disconnected No Global Network Connection Dynamically Changing Topology Reactive Routing Data-Centric Unattended Network Uses Multicast to propagate request

Network

Alice

Bob

7DS Network

Discovers 7DS

Multicast Q

uery

Multicast Query

Multicast Query

Retrieved Object

Internet

Proxy Multicast

SMTP ServerCache

System Architecture & Proxy Server

Proxy Server listens to the incoming HTTP Requests Peer’s user client uses localhost proxy server by default

Query Multicast is sent through a Query Listener & Scheduler SMTP Server listens to the incoming messages and dumps them up

to the MTA

Proxy ServerMulticast

Server

Web Server(Mini HTTP)

SearchEngine

SMTP Server

CacheCache

Manager

MTARelay Email

To Next Client

Search Engine Provides ability to query

self for results Searches the cache

index using Swish-e library

Presents results in any of three formats: HTML, XML and plain text

Similar in concept to Google Desktop

Query Multicast Engine Used to actually exchange

information among peers Requesting peer broadcasts

a query to the network Responding peers reply if

they have information Send encoded string with

list of matching items Requesting peer retrieves

suitable information

Email Delivery

7DS enables mobile nodes to discover each other and relay messages behaving as MTA. Each node calculates statistics and keeps track of each outgoing message using a database.

Node Discovery

Zero-Configuration Network On-Demand Publishing and Discovering of Services Connection set up on-demand using zeroconf protocol

Similar to AppleTalk, Microsoft NETBIOS, Novell IPX

AP

Zero Configuration

Wireless Coverage

AP

Zero Configuration

Wireless Coverage

Community Extensions (Proposal)

7DS Access Boxat 116th & Broadway1. Users can

contribute community information

2. Users can search for and read community information

Users can generate and share content in the spirit of Web 2.0

7DS in Cluster Networks Sparse scenario

Heavily partitioned network; opportunistic p2p data sharing

Dense scenario full network connectivity; multihop routing for

communication Cluster network

A cluster is an isolated island disconnected from the world

Nodes within a cluster connected by multihop routes Network consists of multiple clusters Likely scenario since nodes are heterogeneous

distributed Context: Email delivery application

Should we incorporate multihop forwarding to 7DS?

A Snapshot of a Cluster Network

AP

Route existsto connectto AP

No route toconnect toAP

Mean Cluster Size E[Cu(n)]

• E[Cu(n)]=A exp (B ) denotes # neighbors• Least square fitting

A=0.9694; B=0.9992• Mean cluster size

exponentially related to mean # neighbors

• Percolation theory shows that many metrics are bounded by exponential function of node density– We have identified

bound is (almost) exact for E[Cu(n)]

Small variance of sample meanof cluster size

2000 simulationsn=200 nodesuniformly distributed

Small variance of sample meanof cluster size

Email Delivery Application If multihop route discovery fails

to find AP, i.e. Pc <1, it is likely <4 mean cluster size < e=55

If route discovery fails to find AP, it is likely cluster size is small

Flooding cluster with replicas is justified

Overhead for finding cluster boundary using MST is also small

Always perform route discovery to find route to AP for immediate email delivery

If no route is found, SRC node creates replicas according to message replication schemes

Pc: Prob. of connecting to APnAP: #APs (nAP« n) n: # nodes (n=200)

3 Message Replication Schemes

BoundaryNodes at cluster boundary are more likely to meet an AP

Discover cluster boundary using MST or Dijkstra shortest path algorithm

Gossiping Each node forwards amessage with some prob.

No boundary discovery

Most replicas are close to SRC, not boundary inefficient

Random WalkSource node creates m replicas

Tx node deletes thereplica after successful transmission

# replicas independent of cluster size

Conclusions

• 7DS makes transparent data exchange, even in absence of Internet, possible

• Data Propagation through and out side of the local network– By new nodes joining and others leaving 7DS Network.

• No user intervention unless absolutely necessary

• New step in practical, large-scale wireless networking with gadgets?– Remains to be seen