Dynamic Team Formation on Dynamic Team Formation on Team-Oriented MulticastTeam-Oriented Multicast
Yiguo Wu and Siavosh Bahrami
Tutor: Yunjung Yi
Prof. Mario GerlaCS 218 Advanced Computer Networks
12/05/2003
OutlineOutline
LANMARPrevious work: Dynamic team formationOur modified dynamic team formation
protocolPreliminary resultsDiscussion
LANMARLANMAR
MANET challenge: scalable routing Hierarchical Routing Protocol
– Assumption: “group affinity model”– Benefit: scalable routing– Limitation: assumes logical subnets known in advance
Logical SubnetLogical Subnet
LandmarkLandmark
Dynamic Team FormationDynamic Team Formation Xiaoyan Hong ,Mario Gerla, " Dynamic Group Discovery and Routing in Ad Hoc Networks, "
In Proceedings of Med-Hoc-Net 2002.
No subnet assumption Two Phases:
– Phase 1) Initialization: traveling companion discovery, K-Hop clustering (connectivity, lowest ID, heuristics), and initial elections
– Phase 2) Maintenance: Hysteresis rule for stable landmark election Phase 1 stability issue?
Stability ProblemStability Problem
C
B
A
Rest of Network
B elects C as landmark
Propagation of inaccurateinformation
A initially elects B
Confirmation & Advertisement Confirmation & Advertisement
C
B
A D
Rest of Network
B elects C as landmark
Propagation of accurateinformation
A elects the next lowest ID
Node B: Failed Elect
Node B: Failed Elect
Node B: Failed Elect
Node B: Failed Elect
Protocol Algorithm Details:Protocol Algorithm Details:Routing Table ExchangeRouting Table Exchange
Exchange local routing tables with direct neighbors periodically
How frequently should we exchange local routing tables?
Protocol Algorithm Details:Protocol Algorithm Details:Companion IdentificationCompanion Identification
Identify Companions- Routing Table entry not “stale”- Has been present in local routing table for at
least time t (avoid roaming nodes)
Protocol Algorithm Details:Protocol Algorithm Details:ElectionElection
Elect Clusterhead– Lowest ID Companion– Pros:
Stable Simple Fast
– Cons: Most likely not close to the minimum dominating set (optimal), i.e. many small clusters that may require further merging
– Note: in current implementation, every election is independent of the previous one raising stability issues (add maintenance heuristics?)
Protocol Algorithm Details: Protocol Algorithm Details: Close Proximity MergeClose Proximity Merge
Small cluster problem- Clusters overlap each other
Solution: Low Cost Merging- Observation: If landmarks can see each other
directly, then their corresponding clusters overlap.
- Action: Merge them !
Protocol Algorithm Details:Protocol Algorithm Details:Confirmation & AdvertiseConfirmation & Advertise
Confirmation– Tentatively elected node can “respond” in 2 ways:
If it is its own landmark, or has not elected anyone yet, implicitly accepts.
If elected another node as landmark, rejects by broadcasting failure.
Advertise– If candidate accepts election, then broadcast affiliation
to entire network.
Protocol HighlightsProtocol Highlights
Routing Table Exchange & Companion Identification– Companions are chosen based on the duration
and continuity of their presence within source’s K-hop range
Confirmation and Advertisement periodically
Node type capability also implemented
Implementation Details:Implementation Details:Asymmetric BroadcastAsymmetric Broadcast
When a node casts its vote for landmark, it broadcast the message to compensate for high mobility.
Do we really need to broadcast to everyone?Not really, only the presumed landmark
needs the information.
Implementation Details:Implementation Details:Asymmetric Broadcast cont.Asymmetric Broadcast cont.
“Symmetric Broadcast”High overhead
K-hop range
Presumed Landmark
Implementation Details:Implementation Details:Asymmetric Broadcast cont.Asymmetric Broadcast cont.
“Asymmetric Broadcast”Intermediate nodes only forward
the packet if they can deliverit within TTL.
K-hop range
Presumed Landmark
DemonstrationDemonstration
Qualnet 3.6Provided mobility trace file
– 150 nodes– 6 types (2 mobile)– Terrain size: 1500 x 1350– Duration: 250 seconds
1-Hop clustering
Preliminary Results 1Preliminary Results 1
Same mobility settings as the demonstration
K 100s 150s 200s 250s
1 16 16 13 13
2 11 11 10 10
3 8 8 6 6
Number of LandmarksNumber of Landmarks
Preliminary Results 2Preliminary Results 2
No mobility traces yet, so assumed static model
Tested on 200, 500, 1000 nodes with terrain sizes 20002, 35002, and 60002, respectively
Different K-values: 1 to 31 TypeDuration 200 seconds
Preliminary Results 2 cont.Preliminary Results 2 cont.
K 200 nodes 500 nodes 1000 nodes
1 11 21 66
2 10 14 29
3 2 6 35
Number of LandmarksNumber of Landmarks
Future WorkFuture Work
Cluster maintenance (join, leave, hysterisis rules…)
Parameter tuning– Empirically determine optimal/robust routing table
exchange and election schedules Adaptive K Landmark supergroups: grouping of same-group
landmarks Distibuted/redundant DNS for landmark members
ReferencesReferences
[1] Yunjung Yi, Mario Gerla, Joon-Sang Park, Dario Mazzorri, " Team-oriented Multicast: A Scalable Routing Protocolfor Large Mobile Networks, " NGC 2003. [2] Yunjung Yi, Xiaoyan Hong, Mario Gerla, " Scalable Team Multicast in Wireless Ad hoc Networks Exploiting Coordinated Motion, " NGC 2002. [3] Xiaoyan Hong ,Mario Gerla, " Dynamic Group Discovery and Routing in Ad Hoc Networks, " In Proceedings of Med-Hoc-Net 2002. [4] Wu, F. Dai, M. Gao and I. Stojmenovic, " On calculating power-aware connected dominating sets for efficient routing in ad hoc wireless networks, " IEEE/KICS Journal of Communication Networks, Vol. 4, No. 1, March 2002, 59-70.[5] G. Pei, M. Gerla, and X. Hong, " LANMAR: Landmark Routing for Large Scale Wireless Ad Hoc Networks with Group Mobility," In Proceedings of IEEE/ACM MobiHOC 2000, Boston, MA, Aug. 2000. [6] J.T. Tsai and M. Gerla, “Multicluster, Mobile, Multimedia Radio Network”, ACM/Kluwer Journal of Wireless Networks, Vol.1, No.3, pp. 255-65, 1995.