1. 4.5 Multicast Communication
From Distributed Systems: Principles and Paradigms
Presented by: Daniel Burkhart

2. Contents What is multicast communication?
Application-level multicasting
Gossip-based data dissemination
Current Research Works
Future Research Direction

3. What is multicast communication?
“the support for sending data to multiple receivers.”
belonged to the domain of network protocols, where numerous proposals for network-level and transport-level solutions have been implemented and evaluated.
A major issue in all solutions was setting up the communication paths for information dissemination. In practice, this involved a huge management effort, in many cases requiring human intervention
With the advent of peer-to-peer technology, and notably structured overlay management, it became easier to set up communication paths

4. What is multicast communication (cont’d)
As peer-to-peer solutions are typically deployed at the application layer, various application-level multicasting techniques have been introduced.
Multicast communication can also be accomplished in other ways than setting up explicit communication paths. gossip-based information dissemination provides simple (yet often less efficient) ways for multicasting.

5. Application-level multicasting
The basic idea in application-level multicasting is that nodes organize into an overlay network, which is then used to disseminate information to its members.
An important observation is that network routers are not involved in group membership.
As a consequence, the connections between nodes in the overlay network may cross several physical links, and as such, routing messages within the overlay may not be optimal in comparison to what could have been achieved by network-level routing.

6. Application-Level: Overlay Network
First, nodes may organize themselves directly into a tree, meaning that there is a unique (overlay) path between every pair of nodes.
An alternative approach is that nodes organize into a mesh network
which every node will have multiple neighbors
there exist multiple paths between every pair of nodes.
The main difference between the two is that the latter generally provides higher robustness: if a connection breaks because a node fails, there will still be an opportunity to disseminate information without having to immediately reorganize the entire overlay network.

7. Tree Based Overlay Network – ALM in Chord
Initiator generates a multicast identifier mid.
Lookup succ(mid), the node responsible for Mid.
Request is routed to succ(mid) which will become the root.
If P wants to join, it sends a join request to the root.
When request arrives at Q,
Q has not seen a join request before => it becomes forwarder, P becomes a child of Q. Join request continues to be forwarded.
Q knows about tree => P becomes child of Q. No need to forward join request anymore.

8. App Level Multicast: Building an efficient Tree?
A – root of the multicast tree.
A multicasts a message to other nodes:
Message traverse <B,Rb>, <Ra,Rb>, <Rc,Rd>, <D,Rd> twice.
Can we improve this?

9. Building an efficient tree (2)
Instead of overlay link B to D; we need to form it from A to C.
Double traversal of <Ra,Rb>, <Rc,Rd> would be saved.

10. ALM Quality Measurement
Link stress: How often does an ALM message cross the same physical link?
For example a message from A to D needs to cross (Ra, Rb) twice.
Stretch: Ratio in delay between ALM- level path and network-level path.
For Example Messages B to C follow path of 59 units at ALM but -+7 at network level => Stretch 1255.

11. Finally, The tree cost
The tree cost is a global metric generally related to minimizing the aggregated link costs.
For example, if the cost of a link is taken to be the delay between its two end nodes.
then optimizing the tree cost boils down to finding a minimal spanning tree in which the total time for disseminating information to all nodes is minimal

12. Multicast parental problem?
Assume that a multicast group has an associated and well-known node that keeps track of the nodes that have joined the tree.
When a new node issues a join request, it contacts this rendezvous node to obtain a list of members.
Our goal: select the best member that can be this new node’s parent in the tree.
Who would it select????
There are many alternatives and different proposals follow different solutions.

13. Parental Solution 1 – single source
Consider for example a multicast group with a single source.
This proposal believes that the selection of the best node is always the source.
Why?
Because in that case we can assure that the stretch is 1.
However it would introduce a star topology with the source in the middle.
Although simple, the source would easily become overloaded - only those nodes may be chosen who have K or less neighbors with K being a design parameter.
A star topology is a topology for a Local Area Network (LAN) in which all nodes are individually connected to a central connection point, like a hub or a switch. A star takes more cable than e.g. a bus, but the benefit is that if a cable fails, only one node will be brought down.

14. Parental solution 2 – Switch Trees
Written by Helder and Jamin in 2002.
Assume we have a multicast tree with a single source as root.
Node P can switch parents by dropping the link to its current parent in favor of a link to another node.
Constraint 1: The new parent can never be a member of the subtree rooted at P (because it would create a loop).
Constraint 2: The new parent will not too many immediate children (needed to limit the load of forwarding messages by any single node).

15. Deciding to switch parents?
There are different criteria for deciding to switch parents.
Criteria 1: Optimize the route to the source, effectively minimizing the delay when a message is to be multicast.
To this end, each node regularly receives information on other nodes.
At that point the node can evaluate whether another node would be a better parent in terms of delay along the route to the source, and if so, initiates the switch.
Criteria 2: the delay to the potential other parent is lower than the other parent.
If every node takes this as criterion then the aggregated delays should be minimal.
However, more information would be needed to construct such a tree.

16. An important problem – node failure
In switch trees a simple solution is proposed.
Whenever a node notices that its parent has failed, it simply attaches to the root.
After attachment, the optimization protocol can proceed as usual and will eventually place the node at a good point in the multicast tree.
Helder and Jamin show that the resulting tree is close to minimal spanning.

17. Gossip-based data dissemination
An increasingly important technique for disseminating information is to rely on epidemic behavior.
The main goal is to rapidly propagate information among a large collection of nodes using only local information.
There is no central component by which information dissemination is coordinated.
In this scenario, we assume that all updates for a specific data item are initiated at a single node, avoiding write-write conflicts.

18. How does it work???
Update operations are performed at a single server.
A replica passes updated state to only a few neighbors.
Update propagation is lazy, i.e., not immediate.
Eventually, each update should reach ever replica.
Infect all nodes with the new information as fast as possible.

19. Two main epidemic models
Anti-Entropy
In this model, a node P picks another node Q at random, and subsequently exchanges updates with Q.
Rumor-Spreading (gossiping)
If node P has just been updated for data item x, it contacts an arbitrary other node Q and tries to push the update to Q.
However, it is possible that Q was already updated by another node. In that case, P may lose interest in spreading the update any further, with probability 11k. It then becomes removed.

20. Anti-entropy Details There are three approaches to exchanging updates:
P only pushes its own updates to Q.
P only pulls in new updates from Q.
P and Q send updates to each other (a push-pull approach)

21. Problems with anti-entropy?
In a push only approach – updates can ONLY be propagated by infected nodes.
However, if many nodes are infected, the probability of each one pf those selecting a susceptible node is small.
Also, chances that a particular node remains susceptible for a long period of time because it is not selected by an infected node.
The pull-based approach works better when many nodes are infected since updates are triggered by susceptible nodes.
However, if only a single node is infected push-pull remains the best strategy (jelasity et al., 2005a).

22. Gossiping
Gossiping cannot guarantee that all nodes will be actually updated.
It can be shown that when there is a large number of nodes that participate in the epidemics the fraction s of nodes that will remain ignorant of an update that is remain susceptible satisfies the equation.
𝑠= 𝑒 −(𝑘+1)(1−𝑠)

23. Gossiping (2)
One of the main advantages of epidemic algorithms is their scalability due to the fact that the number of synchronizations between processes is relatively small compared to other propagation models.
Spreading the deletion of a data item is hard.
Deletion of a data item removes all information on that item - so that when old copies of the data item are received, it will be perceived as a new update.
Record the deletion of the data item is just another update – keep a “death Certificate.”

24. Gossiping (3) – deleting values
Problem with death certificates? – they should eventually be cleaned up.
Solution?
Demers et al 1987 – propose that each death certificate is timestamped when it is created.
If it can be assumed that updates propagate to all nodes within a known finite time, then death certificates can be removed after this maximum propagation time has elapsed.

25. Applications of epidemic protocols
Perhaps the most widely-deployed application is the spreading of updates.
Aggregating information:
Let every node I maintain a variable xi. When I contacts node j they each update their value as:
Obviously after this exchange both I and J will have the same value, being an average of all initial values.

26. Textbook summary
The basic idea is to disseminate information from one sender to multiple receivers.
First, multicasting can be achieved by setting up a tree from the sender to the receivers.
In a peer-to-peer system, solutions have also appeared to dynamically set up trees in a decentralized fashion.
Another important class of dissemination solutions deploys epidemic protocols. These protocols have proven to be very simple, yet extremely robust.
Apart from merely spreading messages, epidemic protocols can also be efficiently deployed for aggregating information across a large distributed system.

27. Current research in multicasting (1)
Energy Efficient Multicasting in Cognitive Radio Networks
Abdullah M. Almasoud and Ahmed E. Kamal.
In this paper, They study the problem of energy minimization for multicasting in cognitive radio networks, using omnidirectional and directional antennas.
Their objectives were to build the multicast routing tree and schedule the secondary users’ transmissions such that the total energy consumption is minimized.
Saw the problem as a Multi-Layer Hyper Graph, then formulates the optimization problem as Mixed Integer Linear Program (MILP).
proposes a heuristic algorithm to solve our problem in polynomial time.
The simulation results shows that the proposed algorithm achieves close performance to the optimal solution.

28. Current research in multicasting (2)
Non-Elementary Routes for Multicasting in Transparent Optical Networks - Miklós Molnár
demonstrated that in sparse splitting TONs (in WDM networks but also in EONs), the restricted condition for looking only for light-trees is useless.
In some cases, non-elementary light hierarchies can solve the routing problem and offer better routes. It is not necessary to be limited to trees. Since trees are particular hierarchies, they correspond to the solution when non-elementary routes are not needed.
In p-cycles based protections, the non-elementary cycles can also be advantageous. They can protect against branching node failures even if node encircling elementary cycle does not exist.

29. Current research in multicasting (3)
programming Abstractions for Wireless Multicasting in Software-Defined Enterprise WLANs - Estefanfa Coronado, Roberto Riggio, Jose Villalon and Antonio Garrido.
The increasing demand for multimedia content and for live broadcasting is bringing renewed interest in multicast applications.
However, multicast over Wi-Fi poses several challenges including low-data rates and coexistence issues with regard to other unicast streams.
Software Defined Networking (SDN) has recently emerged as a novel approach to network control and management.
In this paper They present a novel SDN-based solution for multicast rate-adaptation in Wi-Fi networks. The solution builds upon a new abstraction, named Transmission Policy which allows the SDN controller to reconfigure or replace a certain rate control policy if its optimal operating conditions are not met.
An experimental evaluation carried out over a real-world testbed Showed that this approach can deliver an improvement of up to 80% in terms of channel utilization compared to legacy multicast.

30. Current research in multicasting (4)
Messages propagation Control in Delay Tolerant Networks under Epidemic routing protocol - Ahmed El Ouadrhiri, Mohamed El Kamili and Imane Rahmouni
Delay Tolerant Network (DTN) has the potential to connect nodes in an environment where an end- to-end path may not be available between source and destination.
In this paper, They wanted to transmit an information (message) from a source to a group of people in the network area. Using the epidemic routing protocol the message can reach its destinations as fast as possible, but the protocol consumes too much resources.
They model the messages propagation under the epidemic routing protocol by an Ordinary Differential Equation (ODE).
derive the optimal retention of a message by a node, taking into account that all nodes are infected.
Simulations results show that, using the proposed optimal control in epidemic routing context, they reach the same performances of epidemic routing while minimizing the resource consumption.

31. Predictions
Multicasting will become more and more prevalent in Wireless sensor networks.
A rise in epidemic protocols Use.
New forms of multicasting will emerge that is not contained to trees.
High Energy consumption will no longer be a problem.

32. References
Tanenbaum, Andrew S., and Maarten van Steen. Distributed Systems: Principles and Paradigms. 2nd ed., Maarten Van Steen, 2016.
Almasoud, Abdullah M., and Ahmed E. Kamal. “Energy efficient multicasting in cognitive radio networks.” 2017 IEEE International Conference on Communications (ICC), 2017.
Ouadrhiri, Ahmed El, et al. “Messages propagation control in Delay Tolerant Networks under epidemic routing protocol.”  th International Wireless Communications and Mobile Computing Conference (IWCMC), 2017.
M. Molnár, "Non-elementary routes for multicasting in Transparent Optical Networks,"  th International Conference on Transparent Optical Networks (ICTON), Girona, Spain, 2017, pp. 1-4.
E. Coronado, R. Riggio, J. Villalón and A. Garrido, "Programming abstractions for wireless multicasting in software-defined enterprise WLANs," 2017 IFIP/IEEE Symposium on Integrated Network and Service Management (IM), Lisbon, 2017, pp