1. October 8, 2004MPLS: TE and Restoration1 MPLS: Traffic Engineering and Restoration Routing Basics Zartash Afzal Uzmi Computer Science and Engineering Department Lahore University of Management Sciences

2. October 8, 2004MPLS: TE and Restoration2 Outline Background IP Routing and related problems MPLS Routing Basics Labels and label switched paths Traffic Engineering Restoration Routing Our Research Conclusions

3. October 8, 2004MPLS: TE and Restoration3 Application Scenario A service provider (ISP) with several points of presence (PoPs) geographically distributed ISP provisions applications with “strict” network requirements (e.g., VoIP service) Two major requirements: Guaranteed minimum bandwidth between a source and a destination Less then 50ms recovery time in the event of any network element failure

4. October 8, 2004MPLS: TE and Restoration4 Traditional (IP) Routing Characterized by best effort service Individual nodes (routers) take routing and forwarding decisions Usually based on a pre-computed shortest path Forwarding is destination based When routers forward packets, they only look at the destination address May lead to congestion in some parts of the network

5. October 8, 2004MPLS: TE and Restoration5 IP Routing Example Packet 1: Destination A Packet 2: Destination B S computes shortest paths to A and B; finds D as next hop Both packets will follow the same path Leads to IP hotspots! Solution? Try to divert the traffic onto alternate paths 11 12 A B C A B S D

6. October 8, 2004MPLS: TE and Restoration6 IP Routing Example Increase the cost of link DA from 1 to 4 Traffic is diverted away from node D A new IP hotspot is created! Solution(?): Network Engineering Put more bandwidth where the traffic is! Leads to underutilized links; not suitable for large networks 14 12 A B C S A B D

7. October 8, 2004MPLS: TE and Restoration7 IP Routing Vs MPLS Traditional IP Routing Multiprotocol Label Switching (MPLS) SD 54 3 21 MPLS allows overriding shortest paths!

8. October 8, 2004MPLS: TE and Restoration8 Routing Along Parallel Paths Idea: Let the source make the complete routing decision; source decides the complete path for each flow How this may be accomplished? Attach a label to the IP packets; let everyone make forwarding decision on that label On what basis should you choose different paths for different flows? Define some constraints and hope that the constraints will take “some” traffic away from the hotspot! Use CSPF instead of SPF (shortest path first)

9. October 8, 2004MPLS: TE and Restoration9 MPLS: Basics How did they route along parallel paths? They did use a label They also decided to use a new label at each hop to save on label space Terminology LSP: Label switched path LSR: Label switch router IP DatagramLabel

10. October 8, 2004MPLS: TE and Restoration10 Mpls Flow Progress LSR1 LSR2 LSR3 LSR5 LSR6 R1R2LSR4 D 1 - R1 receives a packet for destination D connected to R2 R1 and R2 are regular routers D destination

11. October 8, 2004MPLS: TE and Restoration11 Mpls Flow Progress LSR1 LSR2 LSR3 LSR5 LSR6 R1R2LSR4 D 2 - R1 determines the next hop as LSR1 and forwards the packet (Makes a routing as well as a forwarding decision) D destination

12. October 8, 2004MPLS: TE and Restoration12 Mpls Flow Progress LSR1 LSR2 LSR3 LSR5 LSR6 R1R2LSR4 D 3 – LSR1 establishes a path to LSR6 and “PUSHES” a label (Makes a routing as well as a forwarding decision) D destination 31

13. October 8, 2004MPLS: TE and Restoration13 Mpls Flow Progress LSR1 LSR2 LSR3 LSR5 LSR6 R1R2LSR4 D 4 – LSR3 just looks at the incoming label LSR3 “SWAPS” with another label before forwarding D destination 17 Labels have local signifacance!

14. October 8, 2004MPLS: TE and Restoration14 Mpls Flow Progress LSR1 LSR2 LSR3 LSR5 LSR6 R1R2LSR4 D 5 – LSR6 looks at the incoming label LSR6 “POPS” the label before forwarding to R2 D destination 17 Path within MPLS cloud is pre-established: LSP (label-switched path)

15. October 8, 2004MPLS: TE and Restoration15 TE Capability Recap Who establishes the LSPs in advance? Ingress routers How do ingress routers decide not to always take the shortest path? Ingress routers use CSPF (constrained shortest path first) instead of SPF Examples of constraints: Do not use links left with less than 7Mb/s bandwidth Do not use links with blue color for this request Use a path with delay less than 130ms

16. October 8, 2004MPLS: TE and Restoration16 MPLS Routing SD 54 3 21 MPLS allows routing on pre-established paths!

17. October 8, 2004MPLS: TE and Restoration17 IP versus MPLS: Summary In IP Routing, each router makes its own routing and forwarding decisions In MPLS, source makes the routing decision Intermediate routers make forwarding decisions In IP Routing, packets usually follow the SPF In MPLS packets follow the CSPF In IP Routing, restoration takes few seconds In MPLS, restoration can be of the order of 10ms

18. October 8, 2004MPLS: TE and Restoration18 CSPF What is the mechanism? First prune all links not fulfilling constrains Now find shortest path on the rest of the topology Requires some Reservation mechanism Changing state of the network must also be recorded and propagated For example, ingress needs to know how much bandwidth is left on links The information is propagated by means of routing protocols and their extensions

19. October 8, 2004MPLS: TE and Restoration19 Restoration Routing Application of Traffic Engineering

20. October 8, 2004MPLS: TE and Restoration20 Restoration in IP network In traditional IP, what happens when a link or node fails? Information needs to be disseminated in the network During this time, packets may go in loops Restoration latency is in the order of seconds

21. October 8, 2004MPLS: TE and Restoration21 Restoration in MPLS S123D Primary Path Backup Path Path Protection This type of “path Protection” still takes 100s of ms.

22. October 8, 2004MPLS: TE and Restoration22 Restoration in MPLS S123D Primary Path Backup Path Element Local Protection Local Protection takes of order of 10ms

23. October 8, 2004MPLS: TE and Restoration23 Opportunity Cost Fast restoration requires that backup paths are established “in advance” Backup provisioning requires bandwidth reservation along the backup paths Backup bandwidth is taken from the primary bandwidth Fewer primary LSPs can be established Can we do something to avoid “wasting” so much bandwidth in backup paths? Try to share the backup bandwidth!

24. October 8, 2004MPLS: TE and Restoration24 BW Sharing in Backup Paths Assumption: Two primary paths, whose backups are sharing bandwidth, must not fail together Is this assumption realistic? Failure is a low probability event Once failure occurs, new primary paths with new backups are computed Failure of another element in that time is unlikely

25. October 8, 2004MPLS: TE and Restoration25 BW Sharing in Backup Paths Example:- S1D1 S2D2 453 b1 b2 max(b1, b2) = LSR

26. October 8, 2004MPLS: TE and Restoration26 Creation of Backup Paths

27. October 8, 2004MPLS: TE and Restoration27 Types of Backup Paths

28. October 8, 2004MPLS: TE and Restoration28 Backup Paths: Definitions A next-hop (nhop) backup path that spans link(i,j) is a backup path which: Originates at node i Merges with the primary at node j Provides restoration for one or more primary LSPs that traverse link(i,j) when: link(i,j) fails

29. October 8, 2004MPLS: TE and Restoration29 Backup Paths: Definitions A next-next-hop (nnhop) backup path that spans link(i,j) and link(j,k) is a backup path which: Originates at node i Merges with the primary at node k Provides restoration for one or more primary LSPs that traverse link(i,j) and link(j,k) when either: Node j fails Link(i,j) fails

30. October 8, 2004MPLS: TE and Restoration30 Activation Sets When an element fails, a number of backups are activated “simultaneously” Such backups are in the activation set of that protected element Backups is a single activation set can not share the bandwidth Backups in different activation sets may share the bandwidth

31. October 8, 2004MPLS: TE and Restoration31 Activation Set for node j What paths are activated when node j fails? NNhop paths that span link(x,j) and link(j,y) for all x,y Note that a node is protected by nnhop paths only!

32. October 8, 2004MPLS: TE and Restoration32 Activation Set for node j

33. October 8, 2004MPLS: TE and Restoration33 Activation Set for link(i,j) What paths are activated when link(i,j) fails: Nhop path that spans link(i,j) Nhop path that spans link(j,i) NNhop paths that span link(i,j) and link(j,x) for all x not equal to i,j NNhop paths that span link(j,i) and link(i,x) for all x not equal to i,j

34. October 8, 2004MPLS: TE and Restoration34 Activation Set for link(i,j)

35. October 8, 2004MPLS: TE and Restoration35 Providing Protection Suppose link(i,j) is traversed by a new primary LSP with bandwidth demand b A backup path “around” the link(i,j) can either be: Nhop path (if node j is egress) NNhop path (if node j is not egress) In either case, point of local repair (PLR) is node i We are protecting the LSP that traverses the triplet(PLR, facility, MP) PLR is always node i Facility is the entity being protected: link(i,j) or node j MP is either node j or some other node adjacent to node j

36. October 8, 2004MPLS: TE and Restoration36 Providing Protection Let the bandwidth corresponding to previously established LSPs traversing the triplet (PLR, facility, MP) is b old The backup path is recomputed with bandwidth demand b new = b old +b Various computation algorithms can be deployed and have been studied

37. October 8, 2004MPLS: TE and Restoration37 Computing the Backups How much bandwidth can be shared? Depends upon the routing information propagated Aggregate information scenario: F ij : BW reserved on link(i,j) for primary LSPs G ij : BW reserved on link(i,j) for backup LSPs R ij : Residual BW on link(i,j) Link(i,j) will propagate above information Note: total primary BW on link(i,j) is F ij +F ji

38. October 8, 2004MPLS: TE and Restoration38 Computing the Backups When the new backup path is nhop, how much is shareable on link(u,v)? F ij +F ji -b old is the maximum bandwidth that will simultaneously be active with new backup The bandwith shareable on link(u,v) is: S uv = max(0, G uv – (F ij +F ji -b old )) When the new backup path is nnhop, how much is shareable on link(u,v)? Note that nnhop is protecting against a link as well as a node. Thus, the bandwidth required for both the activation sets must be computed Max(F ij +F ji -b old,  F xj -b old ) is the maximum that will simultaneously be active with the new backup The bandwidth shareable on link(u,v) is: S uv = max(0, G uv – max(F ij +F ji -b old,  F xj -b old ))

39. October 8, 2004MPLS: TE and Restoration39 Computing the Backups PLR knows R uv and S uv for all links PLR computes total bandwidth R uv +S uv available to route the new backup path on each link(u,v) All links for which R uv +S uv < b new are pruned For each remaining link(u,v), the additional bandwidth required is given by max(0, b new -S uv ) PLR computes the route that requires minimum additional bandwidth Note: The computed path is sub-optimal

40. October 8, 2004MPLS: TE and Restoration40 Simulation Parameters 20 node ISP network Each link with capacity 120 units 380 possible pairs LSP requests arrive one by one Ingress/Egress chosen randomly Bandwidth demand for each request is uniformly distributed between 1 and 6 Call holding time is infinite 10 experiments with randomly selected ingress/egress pairs and traffic demands

41. October 8, 2004MPLS: TE and Restoration41 Schemes Compared Kini’s scheme Signalled path is suboptimal Reservations made are corrective Facility Optimal path is signalled Static pools for primary and backups NPP Primary and backups dynamically allocated Optimal path is signalled

42. October 8, 2004MPLS: TE and Restoration42 Results

43. October 8, 2004MPLS: TE and Restoration43 Results