Presented by Anish Sunkara & Mahmoud ElGammal Modeling and Analysis of Regional Registration Based Mobile Multicast Service Management

Overview  Introduction  Related Work  Protocol Description  Simulation Model  Performance Evaluation  Conclusions and Future Work

Introduction  Multicasting: delivering data from a single source to multiple receivers.  Results in great cost savings when implemented efficiently: Minimizing data duplication: otherwise, little advantage to multiple connections to the source. Minimizing the distance traveled by each packet: otherwise, results in low QoS.  Challenges faced by mobile multicasting: Dynamic group membership. Dynamic member topology.

 MH: Mobile Host (= leaf node in the multicast tree)‏  MMA: Mobile Multicast Agent (= non-leaf node in the multicast tree)‏  HA: Home Agent  FA: Foreign Agent

Basic Schemes for Mobile Multicasting 1/2: Remote Subscription (RS)‏  MHs have to subscribe to their multicast group whenever they enter or change their foreign network.  Update frequency to multicast tree = hand-off frequency.  Advantages: Data is always delivered on the optimal shortest path.  Disadvantages: High overhead for reconstructing the multicast tree whenever a hand-off occurs. Doesn't handle source mobility.

Basic Schemes for Mobile Multicasting 2/2: Bi- directional Tunneling (BT)‏  Each MH receives its multicast data via unicast from its HA.  Advantages: Handles source mobility as well as recipient mobility. No need to update the muticast tree on each hand-off.  Disadvantages: The routing path for muticast packet delivery may be far from optimal. Must replicate and tunnel multicast packets to each MH regardless of which foreign networks they reside in (limited scalability).

Proposed Solution (URRMoM)‏  User-oriented Regional Registration-based Mobile Multicast.  A user-centric design allowing each MH to determine its optimal MMA service area dynamically.  Combines the advantages of RS and BT.  Attempts to minimize the total network traffic generated due to multicast packet delivery, as well as multicast tree maintenance overhead.

Related work (1/3)‏  Local Registration:  Each MA manages a list of multicast groups that have MHs in its service area.  An MA receives multicast packets for a group, and then tunnels them to FAs that host visiting MHs in the group.  Advantages: Relatively stable structure. Avoids frequent modifications to the multicast tree.  Disadvantages: Doesn't adapt to each MH's mobility pattern, resulting in a high traffic volume. Each MA represents a single point of failure.

Related work (2/3)‏  mMOM (Mobility-based Mobile Multicasting)  MHs apply either RS or BT according to mobility. High mobility -> BT Low mobility -> RS  Every MH must re-register with its FA after a period of residence time.‏  According to whether the FA receives the re-register message or not, the MH's degree of mobility is decided, and either BT or RS is used.  Advantages: Simple and Practical  Disadvantage: Does not allow co-located care-of-address to be used as in Mobile IP.

Related work (3/3)‏  Range-Based Mobile Multicast (RBMoM)‏  Employs a Mobile MA (MMA) to tunnel packets to the FA to which the MH is currently attached.  The information about which MMA is currently serving a MH is recorded at the MH’s HA.  If a MH is out of the current MMA’s service range, then a MMA handoff occurs and another MMA will take over the multicast service for the MH.  Adapts to MH mobility, but incurs unnecessary communication overhead: When an FA subscribes to the multicast tree as an MMA, all other FA in its range have to be notified of this subscription. Each time a MH moves into the area of a FA, it has to query it for the nearest MMA.

 The MMA is responsible for tunneling multicast packets to the FA under which the MH currently resides as long as the FA lies within the MMA's service area.  Each MH has only one MMA at a time with its MMA being changed from time to time as it roams in the network.  Similar to BT, except that a MH receives multicast data by its MMA which changes dynamically instead of from the HA which is static. URRMoM Protocol Description

 The regional service size of a regional MMA is expressed in terms of the number of subnets covered by the regional MMA.  Each MH keeps a counter to record the number of subnets the MH has crossed within the service area of its MMA.  When the MH crosses the boundary of a subnet, the MH will first check if the new FA of the subnet is already in the multicast group (if the FA is itself a MMA of other MHs). URRMoM Protocol Description

 If the current FA is already a MMA for other MHs, the MMA of the MH will be updated to the current FA even though the MH is still within the service area of the last MMA. The counter in the MH will be reset to 0 after a MMA reset.  If the current FA is not in the multicast tree yet, the counter in the MH will be incremented by one in response to the subnet crossing event. Multicast packets would be tunneled from the MMA to the current FA before they are finally forwarded to the MH.

 When the MH moves across the regional service area such that the counter reaches the regional area size (R) and if the new FA is not an MMA itself, the new FA will subscribe to the multicast tree and become a new MMA for the MH.  When all MHs under a MMA leave, the MMA will unsubscribe from the multicast tree.

Types of Moves in RRMoM Intra-RegionalInter-Regional Definition Occurs when a MH performs a location handoff within a multicast service area of a regional MMA. Occurs when a MH moves across a service area (that is, the counter reaches R), thus incurring a multicast service handoff. Action The MMA is changed only if the new FA it enters into is itself a MMA for other MHs. In this case, the MH’s MMA is updated to the current FA. Otherwise, the MH’s MMA remains the same. The MMA will always be changed. If the new FA is itself a MMA, then the MH’s MMA is simply updated to the current FA. Otherwise, the current FA becomes the MH’s new MMA and a multicast tree subscription event is triggered to add the new MMA to the multicast tree.

Finding Optimal R  There exists an optimal service area size that will minimize the network traffic generated due to mobile multicast services.  This value is a function of:  MH mobility and population.  The size and topology of the network.

 The model considers a multicast group with a single source.  The source is a fixed host.  Group memberships don't change, but MHs may roam dynamically from one subnet to another.  M = # MHs in the group.  The network is a square nxn mesh where each node has exactly 4 neighbors.  Each node corresponds to a subnet with a FA.  Each MH can move in any of the four directions randomly with equal probability.

Simulation Model (cont.) Is fixed Each node corresponds to a subnet, each having its own FA. A MH in one node can move freely into any of the four directions connecting it to the other nodes with equal probability.

 Assuming that a MH's residence time in FA is exponentially distributed with mean µ.  It can be shown that the arrival rate of a MH to any FA in our nxn mesh is λ = µ/(n 2 -1)  The arrival/departure process of M MHs to a FA can be modeled as an M/M/∞/M:  When a FA is not serving any MHs, it will subscribe from the multicast tree.

 The probability of a FA not serving any hosts is P 0, which can be calculated as: (1-1/n 2 ) M  The average number of members in the multicast group being served by a single FA can be calculated as:  A MMA on average will cover R subnets, so the average number of multicast members a MMA covers is:  Thus, the number of MMAs in the system is roughly  The probability that a FA that a MH just entered is a MMA (P MMA ) can be calculated as:  The optimal service size R depends on the tradeoff between the multicast group management costs vs. the tunneling cost.

SymbolMeaning Movea timed transition for the MH to move across a subnet areas. MovesMark(moves)=1 means that the MH just moved across a subnet. MH2MMA a timed transition for the MH to inform the current MMA of the CoA change. P MMA probability that the FA that the MH just enters is already a MMA in the multicast tree. Xs Mark(Xs) holds the number of subnets crossed by the MH in a multicast service area. Subscribe a timed transition to inform the multicast source that the current FA will be added into the multicast tree. R service area covered by a MMA (number of subnets covered by a MMA). Releasea timed transition for the MH to claim the current FA as its MMA. Guard:Mark(tmp2)<>0 a guard for transition Release that is enabled if the MH moves to a subnet whose FA is already a MMA. Guard:Mark(Xs)=R a guard for transition Subscribe that is enabled if the number of tokens in place Xs is equal to R.

SymbolMeaning λarrival rate of a MH to any FA in the n by n network. μMH’s departure rate in a FA. Mnumber of MHs in the multicast group. PiPi probability of state i in the queuing network model. λmλm number of join or leave operations to the multicast tree per unit time. λpλp number of multicast packets delivered per unit time. β average number of hops to reach the source for multicast tree subscription/un-subscription. Rnumber of subnets (or FAs) in one service area covered by a MMA. naverage number of MHs in the multicast group in one FA. r sub transition rate of “Subscribe”. qiqi probability of state i in the underlying Markov or semi-Markov model of the SPN model. τ1-hop communication delay in wired networks.

 C Maintenance : Cost incurred per unit time due to control packets for tree management = MMA Subscription cost + MMA Un- subscription cost  C Service = Cost per unit time for delivering multicast packets from the multicast source to MHs in the multicast group.  Goal is to find optimal service area when given a set of parameter values characterizing the operating and workload conditions.  Observations:  Optimal Service size by a MMA: Determined by trade off between C (maintenance) & C (service)  Expect Cm to increase and Cs to decrease as R (Regional Area Size) decreases  As R increases, Cm will decrease and Cs will increase.

 λ m = Number of join or leave operations to the multicast tree/unit time  r sub = Rate at which a member subscribes a new MMA to the multicast tree after it has crossed R subnets  Let β = Average number of hops separating a MMA and multicast source.  Let τ = Average per-hop communication cost.  Total Subscription Rate = r sub x M  Total Un-subscription Rate =

 C Service = Cost for multicast packet delivery, given by the rate at which packets are generated  λ P = Number of multicast packets delivered per unit time.  Number of hops for multicast packet delivery from the multicast source to MMAs is  Number of hops through which packets are tunneled from various MMAs to M MHs is  The steady state probability of state i, q i, 1<= i <= R, needed is solved easily from the SPN model utilizing solution techniques such as SOR or Gauss Seidel.

 Total traffic generated as a function of the service area size R expressed in terms of the number of subnets  Optimal service area size under which the network traffic generated is minimized  As the mesh network becomes larger (as n increases), the optimal service area size becomes larger and larger Cost vs. Regional Area Size (R) with varying n.

 The network traffic generated vs. R as M varies in an 8 by 8 mesh network.  As M increases the optimal R decreases. Cost vs. R with varying number of MHs

 As β increases, the optimal range R increases for the case when M is fixed at 100  Reason, s the distance separating the source and MMA increases, the maintenance cost increases, so the system prefers to have a large service are to reduce the rate of tree subscription/un- subscription operations Effect of Distance between Source and MMA

 A comparison of the network traffic generated due to maintenance vs. the network size n for URRMoM vs. RS and RBMoM at optimizing R values under the same set of parameter values.  URRMoM always produces the least amount of network traffic compared with RS and RBMoM  Reason, RS is just a special case in which R=1, while URRMoM incurs a overhead Comparison of URRM0M vs RS and RBMoM

 An SMPL simulation has been conducted to validate the analytical results reported in Numerical Data and Analysis section.  To ensure statistical significance of simulation results, a batch mean analysis (BMA) technique has been adopted in which simulation period is divided into batch runs with each batch consisting of 2,000 “cost rate” observations for computing an average value.

Cost vs R with varying n

Cost vs R with varying number of MHs

Comparison of URRMoM Vs. RS and RBMoM

 The proposed URRMoM model combines the advantages of RS and RBMoM models.  This approach combines distinct performance advantages of remote subscription and bi-directional tunneling.  The proposed URRMoM system has simpler system requirements & less computation complexity than the RBMoM system.  Effect of Key parameters such are Regional Area Size (R) are provided.  Both Analytical and Simulated results are shown to optimize the service size covered by the MMA.

 Plan to perform empirical validation of the URRMoM system in an experimental testbed.