On Self-Organization in Mobile Ad Hoc Networks Jie Wu Computer and Network Systems Computer & Information Science & Engineering National Science Foundation [email protected]

Table of Contents • • • • •

Self-Organization: Network Science Sample Self-Organized Solutions Broadcasting in MANETs Some Challenging Issues Conclusions

1. Self-Organization: Network Science • Network classification – Wired Networks • LAN, MAN, WAN, and Internet

– Wireless Networks • Infrastructured networks (cellular networks) • Infrastructureless networks (mobile ad hoc networks)

Infrastructureless Wireless Networks • Mobile ad hoc networks (MANETs) • No base station and rapidly deployable • Unit disk graph: host connection based on geographical distance

Network Science • 21st century networking – – – – – – –

Networks enabling the modern life Engineered nets as a major drive Social and biological nets Social networks enabled by telecomm Importance of common engineered networks Exploded interest in network research Network research leading new and growing business

Network Science (cont’d) • 7 major challenges – – – – – – –

Dynamic, spatial location, and info. propagation Modeling and analysis of very large networks Design and synthesis of networks Increasing the level of rigor and math. structure Abstracting common concepts across fields Better experiments and measurements Robustness and security

Self-Organizing Systems • Local view – P2P and simple interaction (mostly local and without sequential propagation)

• Global functionality – Adaptive, robust, and scalable

Self-Organization in Action • Swarm Intelligence – Ants find shortest routes (based on deposited chemical substance called pheromone) • AntNet (Di Caro and Dorigo, 1998) • Ant-Based Control (Schoonderwored, 1997)

Complex Networks • Small world – Six degrees of separation: short path length – Clustered: neighbors of a node being neighbors – E.g., social networks

Complex Networks • Scale-free networks – Node degree distribution: power-law – Many low degree nodes, a few high degree nodes – E.g., Internet and WWW

(cont’d)

Principles and Design Paradigms • Four paradigms (Prehofer and Bettstetter, Comm. Magazine, July 2005)

– – – –

P1: Local interactions with global properties P2: Minimizing the maintained state P3: Adaptive to changes (self-healing) P4: Implicit coordination

II. Sample Self-Organized Solutions • Theory community – Dijkstra’s self-stabilizing system (Dijkstra, 1974) – An illegitimate state (caused by some perturbations) can be changed back to a legitimate state in a finite number of steps

Parallel Processing: FT routing • Safety level

(Wu, 1992)

– Self-organized routing hints – Self-adjustable based on neighbors’ safety levels – Perturbations: faulty links and/or nodes 3

1

3 3 3

1

Mobile Ad Hoc Networks • Self-organized – – – –

Broadcasting (connected dominating set: CDS)* Topology control* Routing (e.g. geographic routing) Cooperation and trust (e.g. game theory)

• Perturbation: switching-on/off and mobility • Existing solutions: P1 & P2

Minimizing the Maintained State • 1-hop information • 2-hop information • 3-hop information

• k-hop information – Discovered via k rounds of Hello exchanges – Topology and other info. – Usually k = 1, 2, or 3

• Neighborhood vs location info.

Self-Organized Broadcasting • Probabilistic and dynamic

(Tseng et al, MobiCom,

1998)

– Counter-, distance-, and location-based

Broadcasting (cont’d) • Deterministic and static: marking process and Rules 1 & 2 (Wu and Li, Dial M 1999) – A node is forwarding node if it has two unconnected neighbors y

x

v

u

w

m

n

Broadcasting (cont’d) • Pruning Rule-k (Dai and Wu, ICC 2003) – Non-forwarding node: if its neighbor set is covered by several connected and higher priority nodes y x v

u

w n m

Self-Organized Topology Control • Probabilistic

(Bough et al, MobiHoc 2003)

• Network connectivity with high probability: if each node connects to its k nearest neighbors

2 u 1

k 3

1-hop neighborhood

Topology Control (cont’d) • Deterministic

(Li, Hou, and Sha, INFOCOM 2003)

• Network connectivity: if each node connects to its neighbors in the local MST (LMST)

u

1-hop neighborhood

Deterministic Solutions • Many deterministic neighborhood- and location-based solutions resort to – Perfect synchrony (fails on P4) – Static network environment (fails on P3)

Hello interval

time

III: Broadcasting in MANETs • Goal – Find a minimum forward node set (CDS) for full delivery

1

• Methods – Static: use a preconstructed CDS – Dynamic: construct a CDS on-the-fly

2

3

4

Graph-theoretic Definition A set in G(V, E) is dominating if all the nodes in the system are either in the set or neighbors of nodes in the set.

Five-Queen Problem

(1850’s)

Sample Broadcasting 1) Only a few nodes are selected as forward nodes 2) Every node receives the broadcast packet 3) forward node set forms a CDS

A Sufficient Condition • Coverage Condition

(Wu and Dai, ICDCS 2003)

– Non-forwarding node: if any pair of its neighbors are connected by higher priority nodes (Known forward nodes have the highest priority)

v

… u

w

Use of Coverage Condition • Timing – static vs. dynamic

• Selection – Self-pruning vs. neighbor-designation

• Space – K-hop information

• Priority – 0-hop (ID), 1-hop (node degree), and 2-hop priority (clustering coefficient)

Special Cases • • • • • • •

MP, Rule 1, Rule 2, & Rule k LENWB (Sucec and Marsic, Rutgers, 2000) SBA (Peng and Lu, CIT, 2000) MPR (Qayyum et al, INRIA, 2000) DP (Lim and Kim, Seoul National Univ., 2000) Span (Chen et al, MIT, 2001) MP + neighbor elimination (Stojmenovic et al, Univ. Ottawa, 2002)

• TDP/PDP (Lou and Wu, FAU, 2003) • ID-based MPR (Adjih et al, INRIA, 2004)

Adaptive to Changes (Self-healing) • Issue: Movement of mobile nodes – A buffer zone is used in existing protocols without having to redesign them (Wu and Dai, INFOCOM 2004)

Adaptive to Changes (cont’d) • Issues: Switched-on/off nodes – Status changes in 1-hop and 2-hop neighbors only (Wu and Dai, I-SPAN 2002)

on/off u

Implicit Coordination • Issues: Asynchronous sampling of local information – Consistent local view • The view of a node is used consistently by its neighbors. • Each view keeps a version by using a timestamp • The timestamp is chosen by the originator of the data packet

Implicit Coordination (cont’d) • Issues: Delays at different stages of the protocol handshake – Use of weak view consistency for conservative decision • Neighborhood status: connected/disconnected • Unstable link: within the k recent “hello” intervals • Adjacent links include all unstable links v

y

v

v

y

y

x

w

x

w

x

w

Extensions • Efficiency – Iterative CDS reduction (Wu, Dai, and Yang, MASS 2005) – CDS reduction using directional antennas (Dai and Wu, TPDS 2005)

• Reliability – K-connected coverage set (Yang, Dai, Cardei, Wu, MASS 2005)

– K-connected K-dominating set (Dai and Wu, IPDPS 2005)

• Other – Replacement path for edges (Yang, Dai, and Wu, ICDCS 2007)

– Applications in broadcasting and sensor-actor networks (Wu, Yang, and Cardei, INFOCOM 2008)

Iterative Local Solution • Iterative application of a local solution • Seamless integration in a dynamic network: dynamic node priority

Directional Antennas • Energy reduction: broadcasting with directional antennas – Partitioned 2-D space: using topology info – Generalized broadcast coverage condition 3 4 40 2 1 3 4 16 2 1

3 4 20 2 1 3 4 49 2 1

Reliability • Exploiting redundancy: K-connected & Kdominated – Gossip-based: forwarding status with probability P – Extended coverage condition: multiple disjoint paths – Color-based: colored sub-graphs through partition

Other Extensions • Edge coverage condition (ICDCS 2007) – Edge (v, w) is unmarked if a directed replacement path exists connecting v to w via several intermediate edges with higher priorities than (v, w) • Efficient broadcasting (INFOCOM 2008) – Using both network coding and directional antennas • Sensor-actor connectivity (INFOCOM 2008) – Different notions of connecting sensors to actors

IV: Some Challenging Issues • Other self-organized solutions – – – –

Self-organized routing Self-organized incentive mechanisms Self-organized cooperation and trust …

Network Topology Design • Unit disk graph as a complex network – Compared with small world, random, and scalefree networks – Unique topological properties to support selforganized solutions in MANETs

Network Topology Design (cont’d) • Different models for network topology – Unit disk graph – Clusterhead graph • Maximum independent set • Constant number of neighbors

– Geometric graph (Yao, Gabriel, and RNG) • Planar graph • Support for position-based greedy routing

– Regular graph (grid, tree, ring, and hexagon)

Competitive Solutions • Problems (and the associated models) with competitive self-organizing solutions – Rule k for CDS in unit disk graphs: probabilistic constant bound – Geometric face routing in planar graphs: guaranteed message delivery

Generic Solutions • Solutions to support four paradigms, especially P3 and P4 – E.g. How to construct correct results from inconsistent local views? – Topology control: LMST

Generic Solutions (cont’d) • Generic solutions using asynchronous views – Each node keeps k recent local views (i.e., k-hop information) View1,View2,…,Viewk – Construct the conservative view, Viewc based on the worst case scenario – Each node makes conservative (and safe) decisions based on Viewc

• A min-max extension to topology control Wu, Globelcom’05)

(Dai and

Extensions to Other Networks • Marking processing and Rule k work well in small-world networks! •

Average node degree: 10

P

CC

l

CDS

0.01

0.96

0.82

1.05

0.02

0.95

0.75

1.08

0.03

0.91

0.7

1.1

• • •

P: l: CC:

percentage of rewiring average path length clustering coefficient

Extensions (cont’d) • Social networks – – – –

Self-organization Hierarchical trust Funds distribution …

Results from Related Fields • Distributed system community – – – –

Global snapshot View consistency Virtual synchrony …

Conclusions • Importance of self-organized design • Basic principles and challenges • Future – Learning from swarm intelligence – What can (cannot) be computed locally – A better (graph) model for dynamic networks

References •

Prehofer and Bettstetter, Selforganization in Communication Networks: Principles and Design Paradigms, IEEE Comm. Magazine, July 2005.

•

J. Wu and F. Dai, Efficient Broadcasting with Guaranteed Coverage in Mobile Ad Hoc Networks, IEEE Transactions on Mobile Computing, May/June 2005.

•

E. Bonabeau, M. Dorigo, and G. Theraulez, Swarm Intelligence:

From Natural to Artificial Systems, Oxford University Press,

1999. •

Network Science, The National

Academies Press, 2005