1. Hierarchical agent-based secure and reliable multicast in wireless mesh networks Yinan LI, Ing-Ray Chen Robert Weikel, Virginia Sistrunk, Hung-Yuan Chung

2. Introduction to Wireless Mesh Networks ● Wireless Mesh Networks (WMN) is a cost effective “last mile” tech generally consisting of o wireless mesh routers (MR), these form the backbone of the network o mesh clients (MC) o gateways for Internet connectivity

3. Intro (cont) Group communication in WMNs have a problem of supporting secure and reliable mobile multicast The paper proposes an efficient algorithm called hierarchical agent-based secure and reliable multicast (HASRM) in order to mitigate this

4. HASRM requirements ● Only authenticated users interface with the multicast group ● Provide forward and backward secrecy ● Must guarantee delivery of packets ● Must support mobile multicast even when they move and change locations / different (MR) areas

5. Integrated mobility, and multicast service management ● The system was designed because: o User mobility can have a significant impact on multicast service management o Performance optimization around egocentric multicast service management may lead to excessive overhead when users are mobile. o Minimizing network cost has significant fringe benefits in regards to the rest of the system

6. HASRM Organization ● Multicast Agents (MA) o an MA is also mesh routers (MR) o responsible for rekeying and group membership management o registers integrated mobility and multicast server management capabilities across other MAs o dynamically determines optimal regional service size in order to reduce over network cost

7. SPN modeling ● SPN is used here to analyze performance o Focuses on the key parameters in HASRM o Under optimal settings outperforms traditional shortest-path multicast algorithms o Also used to model previous work in order to compare results  Including SeGrOM (Secure Group Overlay Multicast), and the paper which this work is extended from

8. Assumptions and design goals ● A multicast group member may join or leave a group at arbitrary times. ● Group members join and leave events can be modeled by a Poisson process with rates of and, respectively. ● There is a p probability of packet loss o It assures packet transmission through a NAK-based retransmission scheme

9. HASRM structure ● Two-levels o Upper level is a backbone multicast tree connecting mesh routers that serve as agents  Tree is updated whenever there is a leave or join event  Tree maintains a list of all routers serving as agents  An agent services a particular multicast group o The lower level / local multicast group and its associated MA  A single MA may contain several MRs  The regional service size is a key parameter with a tradeoff of packet delivery cost and managerial cost  The optimal regional service can be modeled with the optimal threshold of the number of hops a member can be away from its MA ● Referenced Hoptimal ● Non-optimal threshold is referenced by H

10. Secure Key Management ● Members and MAs share a secret key Ku o Established through Diffie-Hellman o Changed when transitioning to a new MA

11. MA Join ● Steps when a MA joins the backbone o Old group key, Kg is discard o New key, K’g is generated by hashing the original key. (i.e.) K’g = h(Kg) o Source sends K’g to the newly joined using public key encryption

12. MA Leave ● Steps when an MA leaves the backbone o Kg needs to be updated by using the key tree approach o Distributes key through PKI to all MAs excluding the one leaving via rekey messages

13. Reliable multicast data delivery ● Straight Forward Procedure o source encrypts the packet using Kg o disseminates the encrypted packet to the subgroups MA through the tree o Each MA decrypts the packet using KG o MA re-encrypts packet with Ku, sends to each group member o Member decrypts using Ku

14. Packet Loss ● When loss is detected from a member o negative acknowledgement (NAK) is sent to MA o MA sends the missing packet to member o After a period of time MA discards packets ● When loss is detected from backbone (via seq num) two options are available o Source multicasts the packet to all MAs o Source sends packet to all MAs who exhibit the loss

15. Packet Loss (cont) ● Local (Lower) layer uses unicast because o Using multicast in a wireless environment can be very costly in a multicast scenario o Eliminates the need for multicast tree maintenance at lower levels o In contrast to using multicast, error correction requires significantly less overhead when dealing with many members

16. Dynamic group membership management (1/5) Member join *MC selects a serving MR *MC -MR communication:

17. Dynamic group membership management(2/5) Member join: MC executes DH protocol & generates a new K u

18. Dynamic group membership management(3/5) Member Leave MA: ● forwards the leave to the source ● removes itself from the backbone if no other client is serviced The source: ● updates the backbone multicast tree ● sends MA the acknowledgement Leave Request Leave Acknowledgement

19. Dynamic group membership management(4/5)

20. Dynamic group membership management(5/5) Mobility Management NEW MR not MA, ● but member of the OLD MA region =>member reports a location update ● not member of the OLD MA serving region=> NEW MR sends join request to backbone multicast tree => become an MA IF NEW MR is MA =>member switches & starts receiving multicast packages ● MC executes DH protocol and generates a new K u

22. Performance Model(1/3) Mobility Rate (σ) 2dim n x n wireless mesh w/wrap around The average unicast path length Markov Chain Model M/M/ ∞ /M (1) P 0 - probability of not servicing any member P 1 - probability that MR services one member

23. Performance Model (2/3) H is the distance threshold avg #MRs covered = 2H 2 -2H+1 2: For any MR and MA 3: 0 Probability MA services exactly one member 4: K multicast scaling factor 5: Leaves on the multicast tree (MAs)

24. Performance Model (3/3) 6: #MRs on the tree 7: Probability that a multicast data packet is delivered to a member H hops away 8: Expected number of retransmissions to a member H hops away 9: Expected hop distance (average length of paths from south to MA) 10: Probability that a multicast packet is successfully transmitted from source to an MA L hops away 11: Expected number of retransmissions to disseminate a packet to an MA

25. Markov Chain

26. SPN Model for HASRM ● SPN for describing a single group member o Token = a location change o Move = the event of member movement o if NEW MR is: MA => transition probability P 1 =1-P MA just MR => 1.transition probability P 2 = P MA 2.the member reports its new location to its MA(trans. MC2MA) 3. MR becomes MA => Reset o After each MC2MA, a token is placed into Hops o When mark(Hops)=H => transition Join is fried. Firing “Join” resets hops from MA to zero

27. SPN Model *mark(P) : number of tokens in place P

28. Costs ● Cost := total #hops ● C s = C S 1 + C S 2 C S 1 : initial multicast and retransmissions to all MAs C S 2 : Weighted cost for retransmissions from MA to a group member C m : Cost of mobility management (15) Cost for security management when leaving or joining a tree (16) Cost for a member to create a new key (17): Cost per leave event (18): Total cost of all operations

30. Performance Evaluation

31. Service to Mobility Ratio ● SMR = λ p / σ ● The average number of the multicast data packets transmitted from the source to a group member during the interval between two serving MR changes of the group number. ● It captures the service and mobility characteristics of group members.

32. Multicast group size and network size

33. γ = M / n 2 γ: Member Population Density

34. HASRM Can Adapt to Changes in γ

35. p, the Loss Probability of Wireless Link

36. HASRM vs. HASRM-S (S: Static) *Let H = 4 for HASRM-S

37. HASRM vs. HASRM-S (cont.)

38. Comparison: HARSM vs. SPT ● Comparison of HASRM and traditional multicast algorithms based on shortest-path tree (SPT) ● the moderate γ ● The total communication cost is per member per time unit metric

39. HARSM vs. SPT (cont.) ● When p is high, SPT performs poorly.

40. Comparison: HASRM vs. SeGrOM ● Secure Group Overlay Multicast ● hierarchical decentralized multicast Algorithm ● SeGrOM Selects a coordinator for each subgroup of group members connected to the same MR. ● Coordinators are similar to MAs. ● The service area of a coordinator is exactly the coverage area of an MR.

41. HASRM vs. SeGrOM (cont.) ● The total communication cost is per member per time unit metric

42. HASRM vs. SeGrOM (cont.) ● When SMR is small (i.e., the mobility rate is high), the figure shows that HASRM copes well with high group member mobility.

43. Conclusion ● HASRM minimizes the overall communication cost. ● Dynamically maintains MAs. ● Dynamically determines optimal regional service size H Optimal.