1. QoS Routing ISDL mhwon@isdl.snu.ac.kr

2. Quality of Service Routing Algorithms for Bandwidth-Delay Constrained Applications Yi Yang, Jogesh Muppala et al

3. Problem Classification Link-optimization routing Link-constrained routing Path-optimization routing Path-constrained routing Link-* is concerned with bandwidth. Path-* is concerned with delay.

4. Previous Related Work Min-Hop WAPF( Widest Path Avaiable Path First ) SWP( Shortest Widest Path ) MIRA( Minumum Interference Routing Algorithm ) SWP & MIRA introduce the concept of ‘ ingress- egress pair ’ to the problem.

5. Attention to MIRA MIRA makes path selection similar to the methods this paper suggests The concept of ‘ Critical Link ’ In routing requests it avoids critical link for future request not to be rejected. It does not guarantee the delay bound.

6. Define Problem Server Based Routing Or Source Routing : Router : Link ( with B and D )

7. Objective Maximize utilization of networks Maximize call blocking ratio Minimize the number of requests rejected

8. DWC( Delay Weighted Capacity ) For some ingress-egress pair Prefer the powerful machine( link ) (B=5,D=3) (5,2) (5,3) (5,4) (5,2) (5,3) (5,1)

9. DWC( cont ’ d ) Is metric used to measure the potential of the network to process some future request This paper suggest the routing method that does not decrease DWC metric (much).

10. Critical Link In DWC link is sorted in order by link-delay. The request is made up of delay-requirement and bandwidth-requirement. The critical link is the bottleneck of the path that limit its bandwidth.

11. Critical Link( cont ’ d ) The set of bottleneck link of In routing some request, do so that critical links are not used.

12. Routing Algorithm MDWCRA ( Maximum Delay-Weighted Capacity Routing Algorithm )

13. Objective of MDWCRA Maximize the weighted sum of DWC of each ingress-egress pair after satisfying the current request Achieve this by 1) determining appropriate weights for the links in the network and 2) route the request along the least weight path.

14. Algorithm Analysis In G(V,E) Graph 1) Dijkstra Algorithm running time O(|E|log|V|) 2) Bellman-Ford Algorithm running time O(|E||V|)

15. Algorithm Analysis ( cont ’ d )  Calculate LP st ( for each in-egress pair O(n log n) )  In doing so, determine the set of critical links ( O(n) )  For a total of p in-egress pair, running-time is O( pn log n )

16. Algorithm Step  Compute DWC value  Compute the set of critical links  Compute the link weights  Eliminate the unavailable link that is less than requested BW  Compute the delay-constrained least-weight path  Route the request from a to b along this delay-constrained least- weight path and update the residual capacities of the network

17. Three Definition ‘ Wl ’

18. Performance Comparison

19. Comparison ( cont ’ d )

21. Routing Bandwidth Guaranteed Paths with Restoration in Label Switched Networks Samphel Norden et al

22. Main Concept Backup Path sharing Maximize backup path sharing using BLD( Backup Load Distrubution ) Matrix We are focused on LSP( Label Switched Path ). This paper emphasize the point that the protocol it suggests can be introduced in current network using OSPF extension.

23. Backup Mechanism  Backup Path with protection  Backup Path with restoration This paper is constrained to 2).

24. Backup Path sharing uv L1 L2 L3 1.LSP#1 is allocated link L1 as primary path and link L3 as backup path 2. LSP #2 is allocated link L2 as primary path and link L3 as backup path 3. LSP #2 can use L3 as Backup path for free.

25. State Variables C L :Link Capacity F L :the bandwidth used to primary path G L :the bandwidth used to backup path R L = C L – ( F L + G L ) This information is maintained in router( or in some central server )

26. Lack of information Assume only the three state variable ( C L, F L, R L ) is maintained uv ij LALA LBLB r1(b1) = 5 r2(b2) = 10 r3(b3) = 12 r new (b new ) = 33 G L B = 28 R L B = 12 P1 P2 P3

27. Lack of information( cont ’ d ) Route new request over LA as primary path Select to route new request over LB as backup path In LB how much amount can be shared?

28. Use of BLDM L1 L2 L3 L4L7 L5 L6 L8 P1 P2 P3 P4

29. Use of BLDM( cont ’ d ) L1 L2 L3 L4L7 L5 L6 L8 B3 B2 B4 B1

30. BLDM BLDM[i,j] => How much amount Link j is backed up in Link i

31. Benefit of BLDM L1 L2 L3 L4L7 L5 L6 L8 P1 P2 P3 P4 P5

32. Benefit of BLDM( cont ’ d ) L1 L2 L3 L4L7 L5 L6 L8 P1 P2 P3 P4 P5 B5 P5

33. Benefit of BLDM( cont ’ d ) P5 does not share link in primary path with P2,P5 With BLDM the bandwidth of B2,B5 can be shared by B5 Without it, it cann ’ t be shared.

34. Example of BLDM L1L2L3L4L5L6L7L8 F10 826186812 12345678 10826186812 20826 306 410 0 5 0 60 70 8 0

35. Mechanism of BLDM Use OSPF extension Consistency problem arises Repository node is introduced.

36. Free Sharable BW L1 L2 L3 L4L7 L5 L6 L8 P1 P2 P3 P4 Rnew

37. Modeling the link Cost This paper suggests the routing algorithm Until now, explain the core of the algorithm From now, Suggests how to implement this algorithm

38. Modeling the link cost(cont ’ d) Prefer to use free bandwidth(FR). Unless FR can support the requested BW, should use Residual BW(R) Link Cost is modeled following our preference.

39. Routing Algorithm  Two-step algorithm  Iterative or Enumeration based algorithm

40. An Efficient QoS Routing Algorithm for Quorumcast Communication Bin Wang Jeniffer C. Hou

41. Meaning of quorumcast Com. A generalization of multicast communication. In quorumcast com. We term multicast group as quorumcast pool(M). We select quorumcast groups(Q) from quorumcast pool.

42. Objective Construct quorumcast routing tree that spans all the quorumcast members and guarantee the maximum delay from source node s to any node is less than the specified value. It is NP-Complete Problem. So Some heuristic is used

43. Description (D=5,C=4) (5,4) Select n quorumcast group from M quorumcast pool that satisfy some specification.

44. Network Model For any link, link delay and cost is defined As previous paper, This paper describe the network as the G(V,E)

45. Problem Definition subject to Find

46. Data Structures RTD RTC DestNodeMinDelayNextHop DestNodeMinCostNextHopMinDelay- FromNextHop

47. Routing Algorithm(1) First, source node s initiates the route construction process Select a node that satisfy the delay-constraint and have minimum cost. Continues until n nodes is selected or all nodes are marked to be included in tree.

48. SELECTION Data Structure Member Variable  Cost : cost from OnTreeNode to u  OnTreeNode : when included in the routing tree, the node that node u is grafted to  Tag : whether or not node u is on the routing tree.

49. SELECTION(cont ’ d) Initially ‘ Cost ’ is calculated from s ‘ Tag ’ is set to ‘ NO ’ ‘ OnTreeNode ’ is set to s

50. Routing Algorithm(2) After first node is chosen, first node should select the next node to be included in tree Next node is selected that has the minimum cost and the delay less than the specified value Iteratively, update OnTreeNode of SELECTION DS Until all n nodes is selected, this process is repeated

51. Example s A B d3 d2 d1 (1,1) (2,1) (3,3) (1,1) (0.5,0.5) (2,1) (1,2)

52. Loop Detection and Removal The decision centered to Min. Cost and or to Min. Delay can induce loop in routing tree.

53. Dynamic Member Join and Leave