Ad Hoc Routing and Mobility in the Internet Emmanuel Baccelli April 6th 2006

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet MANET scaling to large topologies

Internet & Wireless Since the 1990’s, two telecom. revolutions in “parallel” : Internet Mobile wireless communication (hundreds of millions of users worldwide) Internet access point Recently: desire to merge the advantages of the two scalable, “cheap”, multimedia network wireless mobility, ubiquitous access 1rst step: wireless Internet edge access 802.11 wLAN (popular, cheap) 3G cell services (deploying, expensive) 2nd step: desire to use wireless connectivity beyond edge access with mobile ad hoc networking

Traditional Networking vs Ad Hoc Networking non-directional radio connectivity user devices (hosts) A both user and router E A C B shared access medium D Mobile ad hoc networks: → different characteristics One device type Medium (radio) not fully shared Interferences, less bandwidth Nodes = potential router + host Nodes are mobile routing protocol (ex: OSPF) routers Traditional routing protocols fail New levels of performance needed → new routing mechanisms must be used

The Main MANET Routing Solution Families Proactive routing Nodes maintain routes to all destinations Most protocols: link-state → Adaptation of traditional Internet routing, predictable performance Reactive routing Nodes discover routes only on-demand Based on route request flooding and route reply messaging → Departure from traditional Internet routing, unpredictable performance ROUTE REQUEST (global scope) link state flooding (global scope) S HELLO (local scope) ROUTE REPLY D distribute neighborhood description to all the nodes discover neighborhood

MANET Routing in the Internet Focus on proactive link-state MANET routing easier integration in the Internet Every MANET node potentially generates control traffic Difference with traditional networks (only routers) Overhead reduction mechanisms are needed

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes ►Comparison of optimized flooding schemes Comparison of partial topology schemes MANET integration in the Internet MANET scaling to large topologies

Overhead Reduction: Optimized Flooding Def: Reducing the number of (re)transmissions needed to distribute some information network-wide ►Classical flood: every node retransmits once → network covered, but redundantly (especially when the network is dense) ►Optimized flood: only some nodes retransmit → network efficiently covered - reduced bandwidth use - reduced amount of interferences - reduced number of packet collision this node was covered 4 times

Dominating Set Flooding Algorithm: MPR Flooding MPR (Multi-Point Relay) selection: each node selects a subset its neighbors, covering all the neighbors of its neighbors MPR [P. Jacquet, P. Minet, et al. HiperLAN, 1996] S Flooding: only MPRs retransmit and the network is efficiently covered MPR [L. Viennot, A. Laouiti, A. Qayyum, 2000]

Tree Flooding Algorithm: Gateway Flooding 1. Automatic tree formation: each node selects its neighbor which has the max number of neighbors as parent in the tree R [C. Bonnet, H. Labiod, N. Nikaein, 2000] 2. Network = forest of trees roots = local degree maxima R Gateway Flooding: retransmission happens only along tree vertices (and between trees)

Flooding Algorithms Comparison & Analysis MPR Flooding Simulations (1D) Gateway Flooding Simulations (1D) number of nodes in the area number of retransmitters MPR flooding Gateway flooding classical flooding the proportion of retransmitters is inversely proportional to the number of nodes 1 > 2 3 + o( ) ν the proportion of retransmitters is approx. 70% of the number of nodes Retransmitters % 2 ν ~ → more efficient Node density (ν) [E. Baccelli, P. Jacquet, 2004] → analitycal results confirm this behaviour when the network is dense

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes Comparison of optimized flooding schemes ► Comparison of partial topology schemes MANET integration in the Internet MANET scaling to large topologies

Overhead Reduction: Partial Topology Def: Reducing the amount of routing information that needs network-wide flooding J K P Link state routing can still provide routes with partial link state advertizement I C L O A N MPR G - using only links towards nodes in a connected dominating set (CDS) B H M MPR → route stretching - using only MPR selection links Note: reactive protocols have partial topology as main ideal → no route stretching [P. Jacquet, L. Viennot et al., 2002]

Partial Topology Algorithm Simulations Blue: full link state Purple: CDS links Light blue: MPR links MPR topology reduction is efficient while keeping routes optimal Number of advertized links Number of nodes in the network [E. Baccelli, T. Clausen, P. Jacquet, 2004]

Overhead Reduction Summary MPR flooding and MPR partial topology: simple & efficient MPR techniques are extracts from OLSR (Optimized Link State Routing): OLSR = standard (the IETF’s MANET proactive routing protocol) RFC 3626 [T. Clausen, P. Jacquet, 2003] Link state routing, enhanced with MPR techniques Proven, robust MANET routing solution (up to ~70 nodes with 802.11) →focus on OLSR

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet ►OSPF on wireless Address Autoconfiguration MANET scaling to large topologies

OSPF: Open Shortest Path First OSPF is the Internet’s traditional routing protocol Backbone (Area 0) Area 3 Area 1 Internet routers Area 2 Advantages Scalable: address area hierarchy Generic: modularity for heterogeneous subnetwork tech. Drawbacks Heavy: complexity, overhead Not dynamic: slow to converge, change in the routing → alarm

OSPF Evolution Internet growth: OSPF challenged by: more links more mobile wireless devices OSPF challenged by: More dynamics Wired/wireless heterogeneity →IDEA: design an OLSR-based module for OSPF FACTS: 1. OLSR is close to OSPF (proactive, link state) 2. OLSR is adapted to wireless & dynamics 3. OSPF is modular

Designing OSPF’s Mobile Ad Hoc Module Area 2 Area 1 OSPF shortcomings on wireless: Overhead scalability Database synchronization Acknowledgements Database exchange Cross-area mobility wOSPF C ACK B D A PACKET B retransmits in vain Heavy: new neighbors exchange whole databases Not adapted to wired/wireless heterogeneous synch. needs OSPF control traffic is too big, overhead saturates wireless links Positive ACK fails If A entered a new area, reconfiguration of the node is needed → new address B sends a packet to A A moves [J. Ahrenholz, E. Baccelli et al., IETF 2003] designs a new OSPF extension (called wOSPF) based on OLSR techniques using: - MPR flooding and partial topology - A new database synchronization mechanism

Wired/Wireless Heterogeneity wOSPF: potential mix of large numbers of fixed & mobile nodes in the Internet Heterogeneous needs for synchronization in the network Mobile wireless nodes generate frequent updates (~1 second): synchronization is wasteful Traditional fixed nodes do not generate frequent updates (30min): synchronization is desireable → Need for a new heterogeneous synchronization scheme

Database Signatures (DbX) Synchronization [E. Baccelli, T. Clausen, P. Jacquet, 2004] 1. Each node periodically emits hash of different parts of its link state database (signature) SYNCH A 2. Neighbors compare their respective signatures & detect discrepancies B Signature = ? New node Db exchange Context Independent ACK Signature ≠ 3. Nodes synchronize when discrepancies are detected Light-weight synchronization mechanism, for: Packet loss recovery (replacing ACK) Heterogeneous database synchronization

Analytical Performance Evaluation of DbX traditional OSPF database scanning proportional to n →light weight synch. number of exchanged IP addresses database signatures proportional to log(n) Size of the database (n) size of the network [E. Baccelli, T. Clausen, P. Jacquet, 2004]

wOSPF in a Nutshell wOSPF supports MANET techniques in OSPF framework MPR flooding & partial topology provides efficient overhead reduction DbX provides an efficient database synchronization scheme adapted to the wired/wireless heterogeneity wOSPF OSPF overhead scalability ACK issue solved Wired/wireless synchronization Area mobility

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet OSPF on wireless ►Address Autoconfiguration MANET scaling to large topologies

IP Address Autoconfiguration Prior to routing: nodes should be uniquely identified In the Internet: identity is given by IP address MANET characteristics challenge Internet addressing a “new” node must be configured traditional autoconf. tools fail server(s) may be unreachable no fully shared medium = issue Nodes 123.12.134.53 123.12.134.110 123.12.134.23 MANET Internet 123.12.134.54 123.12.134.45 123.12.134.78 N New node. Address? → Need for new autoconfiguration schemes, adapted to MANETs

Light Autoconf for OLSR [E. Baccelli, T. Clausen, 2005] 1. Configured nodes advertize ADDR-BEACON periodically 2. A new node listens for ADDR-BEACON OLSR Nodes C ADDR-CONFIG 3. A new node registers with a configurating node Proxy Autoconf HELLO N N N NOA-OLSR [K. Mase, C. Adjih, 2005] 4. Temporary address assignment + HELLO tracking Light OLSR Autoconf new OLSR node New Node 5. The configurating node uses OLSR connectivity to act as autoconf proxy Auto. DHCP 6. Permanent address assignment, the new node can take part in the network Manual config Connected with DHCP Connected without DHCP Disconnected with one node configured Partition Merge

Conclusions on MANET Integration in the Internet A suite of new protocols is needed to integrate MANETs in the Internet {wOSPF or OLSR} + autoconfiguration provides a solution for local MANET mobility integration MANET integration standalone MANET connected MANET connected scalable MANET

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet MANET scaling to large topologies ►Introduction to MANET scaling Horizontal scaling with OLSR Fisheye Vertical Scaling with OLSR Trees

Scaling with Mobility in the Internet Internet scaling: tied to IP address location & low mobility Integrating mobile devices: a challenge for Internet scaling MANETs can integrate fixed & wireless mobile devices But MANETs do not scale yet MANET testing (802.11) theoretical OLSR limit practical Internet scalability theoretical size of the Internet 1 ~26 ~212 ~230 ~2128 nodes nodes nodes nodes

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet MANET scaling to large topologies Introduction to MANET scaling ►Horizontal scaling with OLSR Fisheye Vertical Scaling with OLSR Trees

MANET Horizontal Scaling Theory: optimal network capacity is O(√n) Real protocols: network capacity tends to crumble when the numbers of nodes/users increases in the network [P. Gupta, P. Kumar, 1999] OLSR Total network capacity Network population (n) [E. Baccelli, P. Jacquet, C. Adjih et al, 2004] → overhead limits the number of nodes in a MANET

Fisheye OLSR Fisheye: a simple function periodically updating the scope of flooded messages [G. Pei, M. Gerla, T. Chen, 2000] [C. Adjih, E. Baccelli, P. Jacquet et al, 2004] combines OLSR with a simple Fisheye strategy → Fisheye OLSR with Fisheye OLSR routing: network capacity continues to grow with more nodes in the network!!!! S

Analysis of Fisheye OLSR Capacity capacity continues to grow!! with Fisheye OLSR O(√n) Total network capacity (nominal) OLSR Network size (n) [C. Adjih, E. Baccelli, P. Jacquet et al, JCN 2004]

ZOOM: why does Fisheye OLSR scale? Where does √n come from? Fisheye bounding overhead Basic OLSR’s overhead proportional to n Fisheye strategy bounds OLSR’s overhead Similar to Gupta & Kumar minus overhead “tax” Overhead Distance (hop count)

Agenda Background on Mobile Ad Hoc Networks MANET routing overhead reduction schemes MANET integration in the Internet MANET scaling to large topologies Introduction to MANET scaling Horizontal scaling with OLSR Fisheye ►Vertical Scaling with OLSR Trees

MANET Vertical Scaling OLSR Trees MANET Vertical Scaling Fact: the main MANET solutions don't use clustering, hierarchical networking → Clustering and hierarchical routing can scale R [E. Baccelli, 2005] tree-based dynamic clustering (similar to Gateway trees) Hierarchical routing R “super” OLSR between trees (between roots)

Short on MANET Scaling Currently: MANETs are limited in the number of nodes they can manage OLSR + specific scaling techniques (OLSR Trees, Fisheye OLSR) can scale to much bigger topologies Fisheye OLSR + trees scalability 1 26 2128 ? nodes nodes

Perspective on Internet & Wireless ubiquitous wireless Internet We can anticipate: Massive ad hoc topologies Hybrid networks Ubiquitous network traditional Internet fixed core mobile ad hoc nodes ►A new suite of protocols is needed for ubiquitous wireless Internet Research challenges: MANET Scalability MANET Security MANET Multicast MANET Integration

Related Publications IETF Publications: Academic Publications: E. Baccelli, T. Clausen, P. Jacquet, ``Ad-hoc and Internet Convergence: Adapting OSPF-style Database Exchanges for Ad-hoc Networks,'' HET-NET, 2004. C. Adjih, E. Baccelli, P. Jacquet, ``Link State Routing in Wireless Ad Hoc Networks,'' MILCOM, 2003. E. Baccelli, R. Rajan, ``Monitoring OSPF Routing,'' IM, 2001. E. Baccelli, P. Jacquet, ``Diffusion Mechanisms for Multimedia Broadcasting in Mobile Ad Hoc Networks,'' IMSA, 2004. E. Baccelli, ``OLSR Trees: A Simple Clustering Mechanism for OLSR,'' MED-HOC-NET, 2005. C. Adjih, E. Baccelli, T. Clausen, P. Jacquet, R. Rodolakis, ``Fish Eye OLSR Scaling Properties,'' IEEE Journal of Communication and Networks, 2004. IETF Publications: J. Ahrenholz, E. Baccelli, T. Clausen, T. Henderson, P. Jacquet, P. Spagnolo, ``OSPFv2 Wireless Interface Type,'' IETF I-draft, 2004. E. Baccelli, F. Baker, M. Chandra, T. Henderson, J. Macker, R. White, ``Problem Statement for OSPF Extensions for Mobile Ad Hoc Routing,'' IETF I-draft, 2003. E. Baccelli, T. Clausen, R. Wakikawa, ``NEMO Route Optimisation Problem Statement,'‘ IETF I-draft, 2004. E. Baccelli, T. Clausen, P. Jacquet, ``DB Exchange for OSPFv2 Wireless Interface Type,'‘ IETF I-draft, 2004. E. Baccelli, T. Clausen, ``Simple MANET Address Autoconfiguration,'‘ IETF I- draft, 2005.

Thank you.