1. OSPF extension for MANET based on MPR
67th IETF OSPF WG in San Diego, Nov. 7th 2006
OSPF extension for MANET based on MPR
draft-baccelli-ospf-mpr-ext-02.txt
Emmanuel Baccelli

2. Agenda Overview of the extension for MANETs Recent Developments
GTNetS simulation
Next steps

3. OSPF MANET interface type
Addresses mobile ad hoc specificities
Wireless link issues
Highly dynamic topology
MANET interface type properties
Overhead reduction (flooding, adjacency, topology)
Convergence time
OSPF legacy preserved over MANET
MPR approach allows
flooding optimization
adjacency reduction
topology reduction

4. MPR: Multi-Point Relays
What are they?
- subset of links covering the network
MPR Properties:
2-hop neighborhood coverage
shortest paths coverage
MPR
MPR
MPR
... allows for:
- Flooding Optimization
- Adjacency Reduction
- Topology Reduction
Works very well in sparse, dense, slow, fast mobility
(Mathematical models, real world deployments)

5. Recent Developments As discussed in Design Team:
Modified structure to match better RFC 2740 and 2328
Heuristic for MPR selection taking into account link costs
Carried out GTNetS simulations
Inside common framework provided by Boeing
Code base derived from OR code

6. Flooding efficiency with the MPR extension
full LSA
full synchronization
Packet Rates
100 x less
Overhead in Kbits per sec.
1000x1000 square network

7. Flooding Efficiency with the MPR extension
full LSA
full synchronization
Packet Rates
100x less
Overhead in Kbits per sec.
500x500 square network

8. Topology Reduction with the MPR extension
reduced LSA
full synchronization
Packet Rates
similar
Overhead in Kbits per sec.
30% gain
1000x1000 square network

9. Topology Reduction with the MPR extension
reduced LSA
full synchronization
Packet Rates
more packets but
Overhead in Kbits per sec.
50% gain in Kbps
500x500 square network

10. Comparison full LSA versus reduced LSA
in packets/s
in Kbps
75% Gain
in # advertized
links/s
500x500 square network

11. Confirming with Independent simulations (Maple)
Random walk mobility model
full LSA versus reduced LSA
#packets/s
#links/s
500x500 square network

12. Confirming with Independent simulations (Maple)
Random walk mobility model
full LSA versus reduced LSA
#packets/s
#links/s
1000x1000 square network

13. Analytic Model: Why LSA flooding overhead is larger than synchronization overhead
Random walk model, full synchro, full LSA
average link lifetime
links creation frequency (N network size, M average neighbor size)
# nodes ID exchanged in synchro
# links advertized in flooding
: number of retransmissions per neighborhood
: frequency of link creation and dying

14. Next Steps Simulations are necessary step, but: Experimental status
No satisfying wireless model known to date
Scenarii are too restrictive
Not sufficient to evaluate real deployment problems
Experimental status
To leverage the work of the design team so far
To gather real world experience