1. 2007/03/26OPLAB, NTUIM1 A Proactive Tree Recovery Mechanism for Resilient Overlay Network Networking, IEEE/ACM Transactions on Volume 15, Issue 1, Feb. 2007 Zongming Fei, Member, IEEE, and MengkunYang, StudentMember, IEEE Presented by Chih-Yuan Chan

2. 2007/03/26OPLAB, NTUIM2 Abstract Overlay multicast constructs a multicast delivery tree among end hosts. Unlike traditional IP multicast, the non-leaf nodes in the tree are normal end hosts, which are potentially more susceptible to failures than routers and may leave the multicast group voluntarily. Thus, an important problem for making overlay multicast more dependable is how to recover from node departures in order to minimize the disruption of service to those affected nodes.

3. 2007/03/26OPLAB, NTUIM3 In this paper, we propose a proactive tree recovery mechanism to make the overlay multicast resilient to these failures and unexpected events. Each non-leaf node precalculates a parent-to-be for each of its children. When this non-leaf node is gone, all its children can find their respective new parents immediately. The salient feature of the approach is that rescue plans for multiple non-leaf nodes can work together for their respective children when they fail or leave at the same time.

4. 2007/03/26OPLAB, NTUIM4 1. Introduction Overlay multicast (also known as application-layer multicast) implements the multicast functionality at end hosts rather than routers. The study in this paper is resilience to failures and unexpected events. In overlay multicast, a node The key issue is how to reconstruct the overlay tree after these unexpected events. The time to resume the data flow after a node departure.

5. 2007/03/26OPLAB, NTUIM5 When a non-leaf end host leaves the multicast session, all the nodes in the subtree rooted at it are affected. Previous work on tree recovery adopts a reactive approach in which the tree restoration process starts after node departures In this paper, we propose a proactive approach to the recovery of overlay multicast tree.

6. 2007/03/26OPLAB, NTUIM6 First,we need to consider the degree constraints in overlay multicast. Second, we need to be able to deal with multiple departures that happen at the same time. Third, we need to be able to deal with the cases in which the rescue plan is not available at a child.

7. 2007/03/26OPLAB, NTUIM7 2. The Problem Of Restoring Overlay Multicast Trees

8. 2007/03/26OPLAB, NTUIM8 A. Construction of Overlay Multicast Trees The degree-constrained minimum spanning tree problem is an NP-complete problem. We can use heuristics to generate approximate solutions to these problems.

9. 2007/03/26OPLAB, NTUIM9 B. The Problem Of Restoration The focus of this paper is on the problem of restoration of the overlay multicast tree after nodes leave or fail, especially those non-leaf nodes in the tree. Looking at some existing reactive methods which invoke the repair process after node fails. Grandfather Root Grandfather-All Root-All

10. 2007/03/26OPLAB, NTUIM10 3. The Proactive Reconstruction Of Overlay Multicast Trees

11. 2007/03/26OPLAB, NTUIM11 A. Proactive Approach The idea of the proactive approach is to find a rescue plan before the failure happens. For each non-leaf node, it must find parents- to-be for all of its children. Two decisions have been made during the formation of the problem. First, we want to make use of existing parent-child relationships as much as possible, to reduce the impact of the node failure. Second, we want to keep the impact on the grand parent low.

12. 2007/03/26OPLAB, NTUIM12 B. Spanning Tree Problem In general, we can formulate the problem as follows. Suppose node in the multicast tree has n children, as shown in Fig. 2. The total degree, the used degree and the residual degree of node are represented by, and, respectively.

13. 2007/03/26OPLAB, NTUIM13 We set node degree constraint to be For node C j (0 ≦ j ≦ n-1), the degree constraint is In general, in case of, we will find downstream nodes of, specially we will find nodes in the subtree rooted, so that

14. 2007/03/26OPLAB, NTUIM14

15. 2007/03/26OPLAB, NTUIM15 C. Precomputation Algorithm

16. 2007/03/26OPLAB, NTUIM16

17. 2007/03/26OPLAB, NTUIM17 E. Analysis The protocol still works even if the parent-to-be information is not available at a node. First, a node just joins a tree. Its parent has not been able to finish the computation and leaves the tree before informing it of its parent-to-be. Second, it is possible that the parent was too busy with the delivery task and scheduled this background proactive computation for a later time. An observation about the protocol is that it can deal with most multiple failures cases efficiently.

18. 2007/03/26OPLAB, NTUIM18

19. 2007/03/26OPLAB, NTUIM19 F. Discussion on the Applicability of the Proactive Approach The failure recovery problem is formulated for multimedia streaming applications, which impose strict degree constraint on nodes in the multicast tree. The proactive approach can still be used to improve the performance (e.g., the quality of the tree recovered by finding a good contact point for a node with out sacrificing the recovery time.

20. 2007/03/26OPLAB, NTUIM20 4. Performance Evaluation A. Simulation Step B. Responsiveness C. Quality of the Restored Tree D. Recovery Overhead

21. 2007/03/26OPLAB, NTUIM21 A. Simulation Setup The total number of end hosts varies from 80 to 400 in our experiments. application-level distance between two end hosts is the sum of Link latencies on the shortest path between them. The total degree of each node is uniformly distributed between 2 and 6 inclusive.

22. 2007/03/26OPLAB, NTUIM22 B. Responsiveness Responsiveness indicates how fast each scheme can restore the delivery tree after a node fails or leaves the tree. The first metric used for measuring the responsiveness is the average recovery time, which is the average time for an affected node to find a new parent.

23. 2007/03/26OPLAB, NTUIM23

24. 2007/03/26OPLAB, NTUIM24

25. 2007/03/26OPLAB, NTUIM25 The JOIN request sent by a node to its potential parent and/or the reply can be lost. In that case, the node has to retransmit the request, and this can increase the recovery time. Assume the link loss rate at the underlying topology is p link = 1%. The end-to-end path loss rate will be p=1-(1- p link ) l where l is the number of links between a pair of end-hosts. On average l=7.8 and thus the path loss rate p ≒ 8%.The requesting node will wait for some time before the retransmission.

26. 2007/03/26OPLAB, NTUIM26

27. 2007/03/26OPLAB, NTUIM27 C. Quality of the Restored Tree The quality of the restored tree can be measured in two aspects. One is the tree cost, which measures the resource usage of the tree. The other is the maximum delay of the nodes in the tree from the root.

28. 2007/03/26OPLAB, NTUIM28

29. 2007/03/26OPLAB, NTUIM29

30. 2007/03/26OPLAB, NTUIM30

31. 2007/03/26OPLAB, NTUIM31 D. Recovery Overhead For the reactive methods, the control overhead comes from the control messages exchanged for the affected nodes to find new parents. For the proactive scheme, the control messages consists of two parts. 1.Similar to reactive methods, control messages are exchanged for the children of failed nodes to findnew parents, though we may need fewer steps. 2.In addition, every non-leaf node ( except the root ) precalculates parents-to-be for its children.

32. 2007/03/26OPLAB, NTUIM32

33. 2007/03/26OPLAB, NTUIM33

34. 2007/03/26OPLAB, NTUIM34

35. 2007/03/26OPLAB, NTUIM35

36. 2007/03/26OPLAB, NTUIM36 5. Related Work Degree constraints are considered in establishing multicast trees. The proactive approach has been used in recovering link or node failures in multicast tree in the context of the traditional network- layer multicast. Establishing multiple multicast trees is another approach to dealing with failures in overlay multicast. The multiple tree approach can deal with not only the °hard° failure (node failure or leaving), but the °soft° failure (congestion) as well.

37. 2007/03/26OPLAB, NTUIM37 Conclusion Remarks Overlay multicast differs from traditional IP multicast in that the problem of degree constraints is more prominent and non-leaf nodes in the multicast tree are unstable. This paper proposed a proactive approach in which each non-leaf node precomputes the recovery plan for its children. We developed a protocol for nodes to communicate with each other and deal with various failure situations. The recovery process is much faster than the reactive approaches while the quality of the tree and the amortized cost is comparable to those methods.