1. 15-744: Computer Networking L-5 Intra-Domain Routing

2. L -5; 1-29-01© Srinivasan Seshan, 20012 Intra-Domain Routing Routing algorithms Distance vector routing – challenges Link state routing How to make routing adapt to load How to make routing scale Assigned reading [KZ89] The revised ARPANET routing metric

3. L -5; 1-29-01© Srinivasan Seshan, 20013 Factors Affecting Routing 4 3 6 2 1 9 1 1 Routing algorithms view the network as a graph Problem: find lowest cost path between two nodes Factors Static topology Dynamic load Policy D A F E B C

4. L -5; 1-29-01© Srinivasan Seshan, 20014 Outline Distance vector routing – challenges Link state routing Routing metrics Routing hierarchy

5. L -5; 1-29-01© Srinivasan Seshan, 20015 The Bouncing Effect A 25 1 1 B C B C2 1 destcost A C1 1 destcost A B1 2 destcost X

6. L -5; 1-29-01© Srinivasan Seshan, 20016 C Sends Routes to B A 25 1 B C B C2 1 destcost A C1 ~ destcost A B1 2 destcost

7. L -5; 1-29-01© Srinivasan Seshan, 20017 B Updates Distance to A A 25 1 B C B C2 1 destcost A C1 3 destcost A B1 2 destcost

8. L -5; 1-29-01© Srinivasan Seshan, 20018 B Sends Routes to C A 25 1 B C B C2 1 destcost A C1 3 destcost A B1 4 destcost

9. L -5; 1-29-01© Srinivasan Seshan, 20019 C Sends Routes to B A 25 1 B C B C2 1 destcost A C1 5 destcost A B1 4 destcost

10. L -5; 1-29-01© Srinivasan Seshan, 200110 How are These Loops Caused? Observation 1: B’s metric increases Observation 2: C picks B as next hop to A But, the implicit path from C to A includes itself!

11. L -5; 1-29-01© Srinivasan Seshan, 200111 Solution 1: Holddowns If metric increases, delay propagating information In our example, B delays advertising route C eventually thinks B’s route is gone, picks its own route B then selects C as next hop Adversely affects convergence

12. L -5; 1-29-01© Srinivasan Seshan, 200112 Other “Solutions” Split horizon C does not advertise route to B Poisoned reverse C advertises route to B with infinite distance Works for two node loops Does not work for loops with more nodes

13. L -5; 1-29-01© Srinivasan Seshan, 200113 Example Where Split Horizon Fails 1 1 1 1 A When link breaks, C marks D as unreachable and reports that to A and B Suppose A learns it first A now thinks best path to D is through B A reports D unreachable to B and a route of cost=3 to C C thinks D is reachable through A at cost 4 and reports that to B B reports a cost 5 to A who reports new cost to C etc... X B C D

14. L -5; 1-29-01© Srinivasan Seshan, 200114 Avoiding the Bouncing Effect Select loop-free paths One way of doing this: Each route advertisement carries entire path If a router sees itself in path, it rejects the route BGP does it this way Space proportional to diameter

15. L -5; 1-29-01© Srinivasan Seshan, 200115 Loop Freedom at Every Instant Does bouncing effect avoid loops? No! Transient loops are still possible Why? Because implicit path information may be stale See this in BGP convergence Only way to fix this Ensure that you have up-to-date information by explicitly querying

16. L -5; 1-29-01© Srinivasan Seshan, 200116 Distance Vector in Practice RIP and RIP2 Uses split-horizon/poison reverse BGP Propagates entire path Path also used for effecting policies

17. L -5; 1-29-01© Srinivasan Seshan, 200117 Outline Distance vector routing – challenges Link state routing Routing metrics Routing hierarchy

18. L -5; 1-29-01© Srinivasan Seshan, 200118 Basic Steps Start condition Each node assumed to know state of links to its neighbors Step 1 Each node broadcasts its state to all other nodes Reliable flooding mechanism Step 2 Each node locally computes shortest paths to all other nodes from global state Dijkstra’s shortest path tree (SPT) algorithm

19. L -5; 1-29-01© Srinivasan Seshan, 200119 Link State Packets (LSPs) Periodically, each node creates a link state packet containing: Node ID List of neighbors and link cost Sequence number Needed to avoid stale information from flood Time to live (TTL) Node outputs LSP on all its links

20. L -5; 1-29-01© Srinivasan Seshan, 200120 Reliable Flooding When node J receives LSP from node K If LSP is the most recent LSP from K that J has seen so far, J saves it in database and forwards a copy on all links except link LSP was received on Otherwise, discard LSP How to tell more recent Use sequence numbers Same method as sliding window protocols

21. L -5; 1-29-01© Srinivasan Seshan, 200121 SPT Algorithm (Dijkstra) SPT = {a} for all nodes v if v adjacent to a then D(v) = cost (a, v) else D(v) = infinity Loop find w not in SPT, where D(w) is min add w in SPT for all v adjacent to w and not in SPT D(v) = min (D(v), D(w) + C(w, v)) until all nodes are in SPT

22. L -5; 1-29-01© Srinivasan Seshan, 200122 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

23. L -5; 1-29-01© Srinivasan Seshan, 200123 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

24. L -5; 1-29-01© Srinivasan Seshan, 200124 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

25. L -5; 1-29-01© Srinivasan Seshan, 200125 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

26. L -5; 1-29-01© Srinivasan Seshan, 200126 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

27. L -5; 1-29-01© Srinivasan Seshan, 200127 Example AF B DE C 2 2 2 3 1 1 1 3 5 5 B CDEF

28. L -5; 1-29-01© Srinivasan Seshan, 200128 Link State Characteristics With consistent LSDBs, all nodes compute consistent loop-free paths Limited by Dijkstra computation overhead, space requirements Can still have transient loops A B C D 1 3 52 1 Packet from C  A may loop around BDC if B knows about failure and C & D do not X

29. L -5; 1-29-01© Srinivasan Seshan, 200129 Link State vs. Distance Vector In DV send everything you know to your neighbors In LS send info about your neighbors to everyone Msg size: small with LS, potentially large with DV Msg exchange: LS: O(nE), DV: only to neighbors

30. L -5; 1-29-01© Srinivasan Seshan, 200130 Link State vs. Distance Vector Convergence speed: LS: faster – don’t need to process LSPs before forwarding DV: fast with triggered updates Space requirements: LS maintains entire topology DV maintains only neighbor state

31. L -5; 1-29-01© Srinivasan Seshan, 200131 Link State vs. Distance Vector Robustness: LS can broadcast incorrect/corrupted LSP Can be made robust since sources are aware of alternate paths DV can advertise incorrect paths to all destinations Incorrect calculation can spread to entire network

32. L -5; 1-29-01© Srinivasan Seshan, 200132 Link State vs. Distance Vector In LS nodes must compute consistent routes independently - must protect against LSDB corruption In DV routes are computed relative to other nodes Bottom line: no clear winner, but we see more frequent use of LS in the Internet

33. L -5; 1-29-01© Srinivasan Seshan, 200133 Outline Distance vector routing – challenges Link state routing Routing metrics Routing hierarchy

34. L -5; 1-29-01© Srinivasan Seshan, 200134 Importance of Cost Metric Choice of link cost defines traffic load Low cost = high probability link belongs to SPT and will attract traffic, which increases cost Main problem: convergence Avoid oscillations Achieve good network utilization

35. L -5; 1-29-01© Srinivasan Seshan, 200135 Metric Choices Static metrics (e.g., hop count) Good only if links are homogeneous Definitely not the case in the Internet Static metrics do not take into account Link delay Link capacity Link load (hard to measure)

36. L -5; 1-29-01© Srinivasan Seshan, 200136 Original ARPANET Metric Cost proportional to queue size Instantaneous queue length as delay estimator Problems Did not take into account link speed Poor indicator of expected delay due to rapid fluctuations Delay may be longer even if queue size is small due to contention for other resources

37. L -5; 1-29-01© Srinivasan Seshan, 200137 Metric 2 - Delay Shortest Path Tree Delay = (depart time - arrival time) + transmission time + link propagation delay (Depart time - arrival time) captures queuing Transmission time captures link capacity Link propagation delay captures the physical length of the link Measurements averaged over 10 seconds Update sent if difference > threshold, or every 50 seconds

38. L -5; 1-29-01© Srinivasan Seshan, 200138 Performance of Metric 2 Works well for light to moderate load Static values dominate Oscillates under heavy load Queuing dominates Reason: there is no correlation between original and new values of delay after re- routing!

39. L -5; 1-29-01© Srinivasan Seshan, 200139 Specific Problems Range is too wide 9.6 Kbps highly loaded link can appear 127 times costlier than 56 Kbps lightly loaded link Can make a 127-hop path look better than 1- hop No limit to change between reports All nodes calculate routes simultaneously Triggered by link update

40. L -5; 1-29-01© Srinivasan Seshan, 200140 Example Net X Net Y B A

41. L -5; 1-29-01© Srinivasan Seshan, 200141 Example Net X Net Y B A After everyone re-calculates routes:.. Oscillations!

42. L -5; 1-29-01© Srinivasan Seshan, 200142 Consequences Low network utilization (50% in example) Congestion can spread elsewhere Routes could oscillate between short and long paths Large swings lead to frequent route updates More messages Frequent SPT re-calculation

43. L -5; 1-29-01© Srinivasan Seshan, 200143 Revised Link Metric Better metric: packet delay = f(queueing, transmission, propagation) When lightly loaded, transmission and propagation are good predictors When heavily loaded queueing delay is dominant and so transmission and propagation are bad predictors

44. L -5; 1-29-01© Srinivasan Seshan, 200144 Normalized Metric If a loaded link looks very bad then everyone will move off of it Want some to stay on to load balance and avoid oscillations It is still an OK path for some Hop normalized metric diverts routes that have an alternate that is not too much longer Also limited relative values and range of values advertised  gradual change

45. L -5; 1-29-01© Srinivasan Seshan, 200145 Revised Metric Limits on relative change Measured link delay is taken over 10sec period Link utilization is computed as.5*current sample +.5*last average Max change limited to slightly more than ½ hop Min change limited to slightly less than ½ hop Bounds oscillations Normalized according to link type Satellite should look good when queueing on other links increases

46. L -5; 1-29-01© Srinivasan Seshan, 200146 Routing Metric vs. Link Utilization 0 30 60 140 75 50%100%25%75% 225 New metric (routing units) Utilization 9.6 satellite 9.6 terrestrial 56 terrestrial 56 satellite 90

47. L -5; 1-29-01© Srinivasan Seshan, 200147 Observations Utilization effects High load never increases cost more than 3*cost of idle link Cost = f(link utilization) only at moderate to high loads Link types Most expensive link is 7 * least expensive link High-speed satellite link is more attractive than low-speed terrestrial link Allows routes to be gradually shed from link

48. L -5; 1-29-01© Srinivasan Seshan, 200148 Network Response Maps 0.0 1.0 4.0 0.6 0.8 0.4 2.03.01.01.52.53.5 0.5 Link cost Mean load on link 25% 50% 200% Load of “average” link as a function of that link’s cost Created empirically Network load 0.2 400% 100%

49. L -5; 1-29-01© Srinivasan Seshan, 200149 Equilibrium Calculation 0.0 1.0 4.0 0.6 0.8 0.4 2.03.01.01.52.53.5 0.5 Link cost Mean load on link 25% 50% 0.2 HN-SPF D-SPF Combine utilization to cost and cost to utilization maps Equilibrium points at intersections 200%Network load400% 100%

50. L -5; 1-29-01© Srinivasan Seshan, 200150 Equilibrium Utilizations 1.2 9 0.8 1.0 0.6 3612570 0.2 Min-Hop Offered Load Utilization 0.4 0.0 48 Min Hop HN-SPF D-SPF Ideal Equilibrium points vs. offered load

51. L -5; 1-29-01© Srinivasan Seshan, 200151 Routing Dynamics 0 1.0 4.0 0.5 0.75 0.25 2.03.01.01.52.53.5 0.5 Link reported cost Utilization Bounded oscillation Metric map Network response Limiting maximum metric change bounds oscillation

52. L -5; 1-29-01© Srinivasan Seshan, 200152 Routing Dynamics 0 1.0 4.0 0.5 0.75 0.25 2.03.01.01.52.53.50.5 Reported cost Utilization Metric map Network response Easing in a new link

53. L -5; 1-29-01© Srinivasan Seshan, 200153 Outline Distance vector routing – challenges Link state routing Routing metrics Routing hierarchy

54. L -5; 1-29-01© Srinivasan Seshan, 200154 Routing Hierarchies Flat routing doesn’t scale Each node cannot be expected to have routes to every destination (or destination network) Key observation Need less information with increasing distance to destination Two radically different approaches for routing The area hierarchy The landmark hierarchy (discuss in routing alternatives)

55. L -5; 1-29-01© Srinivasan Seshan, 200155 Areas Divide network into areas Areas can have nested sub-areas Constraint: no path between two sub-areas of an area can exit that area Hierarchically address nodes in a network Sequentially number top-level areas Sub-areas of area are labeled relative to that area Nodes are numbered relative to the smallest containing area

56. L -5; 1-29-01© Srinivasan Seshan, 200156 The Area Hierarchy 12 3 1.1 1.2 2.1 2.2 3.1 3.2 2.2.1 2.2.2 1.2.1 1.2.2

57. L -5; 1-29-01© Srinivasan Seshan, 200157 Routing Within area Each node has routes to every other node Outside area Each node has routes for other top-level areas only Inter-area packets are routed to nearest appropriate border router Can result in sub-optimal paths

58. L -5; 1-29-01© Srinivasan Seshan, 200158 Path Sub-optimality 12 3 1.1 1.2 2.1 2.2 3.1 3.2 2.2.1 3 hop red path vs. 2 hop green path start end 3.2.1 1.2.1

59. L -5; 1-29-01© Srinivasan Seshan, 200159 Next Lecture: Inter-Domain Routing Border Gateway Protocol (BGP) Assigned reading [LAB00] Delayed Internet Routing Convergence