Simulative Comparison of MPLS Protection Switching vs. OSPF Re-routing

Transcription

1 C O R P O R A T E T E C H N O L O Y Simulative Comparison of MPLS Protection Switching vs. OSPF Re-routing nformation & Sandrine PASQUALINI Antoine FROT Andreas Iselt Andreas Kirstädter

2 C O R P O R A T E T E C H N O L O Y Introduction MPLS OSPF Label switching Recovery Base protocol New routing algorithm Recovery Simulation CONTENTS nformation & Framework Scenarios Main results Conclusion

3 C O R P O R A T E T E C H N O L O Y Issues : Internet Evolution Increasing amount of multimedia flows over IP Need for Quality of Service (QoS) Objectives : Future and traditional services on a common IP-platform QoS, resilience, scalability Network Control Server Network Admission Control nformation & IP core

4 C O R P O R A T E T E C H N O L O Y Our objectives Simulate the behaviour of a routing protocol : Pre-calculated : Dynamic and distributed : Resilience : MPLS OSPF speed up failure detection improve rerouting Compare the resulting recovery speed

5 C O R P O R A T E T E C H N O L O Y Introduction MPLS OSPF Label switching Recovery Base protocol New routing algorithm Recovery Simulation CONTENTS nformation & Framework Scenarios Main results Conclusion

6 C O R P O R A T E T E C H N O L O Y Attribution of a label Label Switching On ingress router (Label Edge Router) Not only destination based Suppression on egress router Packet forwarding On transient routers (Label Switch Routers) Following a Forwarding Table No analysis of packet s network layer header

7 C O R P O R A T E T E C H N O L O Y nformation & Protection Restoration A A B B MPLS Recovery lobal Regional Local E E H H C C F P1: lobal Protection F R1: lobal Restoration I I D D A A B E H C F P2: Regional Protection (by Haskin) B E H C F R2: Regional Restoration I I D D A A B E H C F P3: Local Protection B E H C F R3: Local Restoration I I D D

8 C O R P O R A T E T E C H N O L O Y Introduction MPLS OSPF Label switching Recovery Base protocol New routing algorithm Recovery Simulation CONTENTS nformation & Framework Scenarios Main results Conclusion

9 C O R P O R A T E T E C H N O L O Y Hello protocol Topology changes detection Link state advertisements (LSA) OSPF Main mechanisms Distribution and storage of the topology Each router maintains its own view Routes calculation Distributed, shortest path Forwarding information base (FIB) Router s architecture Updated when routes have changed T hello detection T dead T lsa T lsaflood 10s 4* T hello propagation 33ms ms T SPT O(n.log n), O(n 2 ) T SPTdelay T SPThold 5s 10s rerouting T FIB ms detection propagation rerouting

10 C O R P O R A T E T E C H N O L O Y Routes convergence Routes convergence time (RCT) failure detection, information s propagation, rerouting 40 50s Possible extensions Reducing T hello, T SPTdelay et T SPThold React faster when resources decrease than when they increase : Multipath routing in order to have at least one backup link When a link fails, react only locally expected RCT < 500ms

11 C O R P O R A T E T E C H N O L O Y "Outdegree 2" Routing (O 2 ) At least 2 outgoing links at each node for each destination No single point of failure No loops, except "jokers" Minimize: paths size/ number of jokers Today s IP-network cores are sufficiently meshed Example : COST 239 "hammoc set" towards Berlin

12 C O R P O R A T E T E C H N O L O Y Speed up failure detection Fault tolerance Reduce delays : T hello =100ms, T SPTdelay et T SPThold =0 Detection at hardware level: 5ms Local reaction (no propagation delays) Change of routing tables Change of distribution weights Preventive mechanism After local reaction is completed, recompute routes

13 C O R P O R A T E T E C H N O L O Y Introduction MPLS OSPF Label Switching Recovery Base protocol New routing algorithm Recovery Simulation CONTENTS nformation & Framework Scenarios Main results Conclusion

14 C O R P O R A T E T E C H N O L O Y OSPF Standard Simulated protocols Accelerated (Hello or hardware detection) Accelerated, with O 2 routing and local reaction MPLS With a detection mechanism equivalent to OSPF Hellos B C B C A E F D A E F D H I H I nformation & P2: Regional Protection (by Haskin) P3: Local Protection

15 C O R P O R A T E T E C H N O L O Y Demand (.ndl) Pathmaker MPLS Paths (.ndl) Ndl2Tcl Topology (.ndl) Tool Chain New routing algorithm Simulation (.tcl) NS-2 output (.nam) NAM

16 C O R P O R A T E T E C H N O L O Y Characteristics COST 239 topology Fully-meshed traffic One link failure Simulation Scenarios Measures Sum of all incoming traffic Mean of values for each possible link failure

17 C O R P O R A T E T E C H N O L O Y User Interface

18 C O R P O R A T E T E C H N O L O Y Just after the failure (2.5s) Overall traffic decreases Similar behaviour but Comparison of SPF / O 2 routing More traffic is affected in O 2 case With equal detection times Equal reaction times Overall traffic in the network [Mbytes] O 2 nformation & Simulation time

19 C O R P O R A T E T E C H N O L O Y Processing of a LSA LSA flooding Routes calculation 0,8 1,20 Restoration time 33 Routes installation (ms)

20 C O R P O R A T E T E C H N O L O Y Comparison of SPF / O 2 routing The time saved thanks to O 2 routing and local reaction is hidden by T FIB Overall traffic in the network [Mbytes] nformation & Simulation time

21 C O R P O R A T E T E C H N O L O Y MPLS protection is always faster Comparison of MPLS / OSPF In general (curve a, standard MPLS) With equivalent detection times (pairs a b, and c d ) 100ms less than OSPF Overall traffic in the network [Mbytes] T FIB nformation & Simulation time

22 C O R P O R A T E T E C H N O L O Y Conclusion Resilience based on traditional OSPF is slow Accelerated OSPF : reaches sub-second recovery improving restoration speed? speed up failure detection but FIB update still an issue MPLS protection has superior resilience : faster recovery but more complex operation, and more bandwidth consuming nformation & Pre-planned LSP restoration with resource sharing

23 C O R P O R A T E T E C H N O L O Y Thank you for your attention Questions?

24 lobal Protection C O R P O R A T E T E C H N O L O Y A Protection switching, pre-established alternate LSP, global scope B E w-lsp p-lsp H C F D + Single backup LSP per working LSP Failure signaling required + Nodefailures covered

25 C O R P O R A T E T E C H N O L O Y Fast Reroute [Haskin] Protection switching, pre-established, pre-reserved, local switching, global recovery Alternative recovery LSP set up from the last-hop LSR in reverse direction to the ingress LSP and along a node-disjoint path to the egress LSP A B E w-lsp H C F p-lsp D + Single backup LSP per working LSP No failure signaling required + Nodefailures covered High spare capacity requireme nformation & Source: [draft-haskin-mpls-fast-reroute-01.txt]

26 Local Protection C O R P O R A T E T E C H N O L O Y Protection switching, pre-established, local scope, pre-reserved A p-lsp1 B E w-lsp H C F p-lsp2 p-lsp3 D Multiple backup LSPsper working LSP + No failure signaling required Nodefailures not covered

27 lobal Restoration C O R P O R A T E T E C H N O L O Y Restoration, established on-demand, reserved on-demand, global scope A B E w-lsp H C F D Failure signaling required + Nodefailures covered + Alternative LSPsdistributed over netwo => high spare capacity efficiency

28 Failure to Egress Restoration C O R P O R A T E T E C H N O L O Y A Restoration, pre-established, pre-reserved, local switching, global recovery C B w-lsp F E D H + No failure signaling required + Nodefailures covered o Between local and global routing => average spare capacity efficien

29 Local Restoration C O R P O R A T E T E C H N O L O Y Restoration, established on-demand, reserved on-demand, local scope A B E w-lsp H C F D + No failure signaling required Alternative LSPs locally routed Nodefailures difficult to cope with => lower spare capacity efficiency