MPLS Recovery Didier COLLE Pim VAN HEUVEN Adelbert GROEBBENS Chris DEVELDER Mario PICKAVET Piet DEMEESTER
MPLS recovery: single layer • Introduction to: – MPLS and MP λ S technologies – MPLS Recovery techniques: • Study of IETF proposals • Development of FTCR scheme • Porting MPLS recovery to MP λ S • Spare resource dimensioning
MPLS and MP λ S IP Payload IP Header A C MPLS Label 7 5 B D IN IF IN LABEL OUT IF OUT LABEL A 2 D 3 B 5 C 7 tributary B 9 D 7 add/drop ports λ OUT λ IN --> λ OUT λ IN OXC aggregate aggregate fiber port fiber port
MPLS protection • Pre-establish backup LSP Backup LSP PML – Protected segment: • local (link or node) • subnetwork Working LSP • end-to-end – Upstream: Protection Switch LSR (PSL) IN IF IN LABEL OUT IF OUT LABEL • protection switching A 1 C 3 B 2 C 3 – Downstream: Protection Merge LSR (PML) • no protection switching, but merging End-to-end Prot A B
MPLS protection: local loop-back Alternative Path A B Alternative Path A B Reuse Alternative path
Re-routing in MPLS Link state update Shortest path Link packets Re-calculations LSP Update the LSP: • O = old next hop N • N = new next hop The Next Hop of S has changed S O B A Link Fails MPLS
FTCR: Fast Topology-driven Constraint-based Rerouting S has topology S updates his S Calculates the knowledge topology new path Link Path Setup Problem: Specify every LSP routing tables not valid hop in path (Explicit routed) network topology N Link Fails S O B A The routing tables will be updated but the MPLS paths are restored before that MPLS
Porting MPLS protection to MP λ S Select (= switch to) best signal IP router IP OTN Working O-LSP Backup O-LSP OXC Select (= switch to) best signal Conclusions : Dedicated, thus 2 wavelengths needed • Dedicated protection • Merging problem • solve by simulating with passive selector/switch • shift merging to client (i.e., IP layer).
Simulations: assumptions • Single layer planning – MPLS recovery techniques – MP λ S recovery techniques • Routing: – shortest path – each LSP independent • C apacity/cost model – linear capacity model: line capacity = used capacity – cost model: cost to carry unit of capacity proportional with link weight (roughly estimated on distance). • T raffic matrices : asymmetric • Random generation (e.g., traffic): – set of 10 instances • MP λ S: wavelength conversion assumed
Results: Optical versus Electrical Recovery Electrical (=Shared) versus Optical (=Dedicated) recovery for the LARGE topology 1000% 900% 800% 700% 600% Relative Cost (spare/working) 500% 400% 300% 200% 100% 0% Optical (=Dedicated) n o R i t Electrical (=Shared) c C g e T n t o F i k t c r u p a n o b o l r a e - i p t c r c o o e l o t o l l r a p c h o t l a p Electrical (=Shared) versus Optical (=Dedicated) recovery for the SMALL topology Failure scenarios: 1000% 900% • single link failures (interpreted as a node failure 800% by adjacent LSRs, except for rerouting) 700% • single node failures 600% Relative Cost Traffic: (spare/working) 500% • Uniform pattern 400% • Randomly generated (integer values) 300% 200% Last link (of an LSP): 100% • Protected 0% Optical (=Dedicated) • Not reverted (for local loop-back) local protection FTCR Electrical (=Shared) rerouting Topologies local loop-back path protection • Large: 57 links and 44 nodes • Small: 36 links and 30 nodes
Results: Optical versus Electrical Recovery • Rerouting and FTCR: no difference – When tearing down part of primary LSP downstream of the failure • Worst case: dedicated versus shared protection – No merging possible (eventually simulating merging via switching) – Label is scarce product in MP λ S, instead of bandwidth in MPLS – How to improve this worst case --> see next slides • Dedicated effect: – significant for end-to-end protection or local loop-back – does not allow sharing between both direction for local loop-back – catastrophe for local protection
Results: Electrical MPLS Recovery GLOBAL versus LOCAL recovery for electrical domain (shared protection) local protection FTCR rerouting working 2 4500 Rerouting: correct view of topology 1.8 4000 FTCR: interprets link as node failure, 1.6 3500 due to hello-msg detection scheme 1.4 Relative cost (spare/working) 3000 1.2 Working cost 2500 1 2000 0.8 1500 0.6 1000 0.4 500 0.2 0 0 LARGE Topology SMALL Topology Path Protection versus Rerouting for electrical domain (shared protection) rerouting local loop-back path protection 1.4 Failure scenarios: • single link failures (interpreted as a node failure 1.2 by adjacent LSRs, except for rerouting) • single node failures 1 Relative cost (spare/working) Traffic: • Uniform pattern 0.8 • Randomly generated (integer values) Last link (of an LSP): 0.6 • Protected 0.4 • Not reverted (for local loop-back) Topologies 0.2 • Large: 57 links and 44 nodes • Small: 36 links and 30 nodes 0 LARGE Topology SMALL Topology
Results: Electrical MPLS Recovery LINE failures for HUBBED demand NODE failures for HUBBED demand 3.3 3.1 3.1 2.9 2.9 2.7 2.7 2.5 2.5 2.3 2.3 2.1 2.1 1.9 1.9 1.7 TO TO 1.5 FROM 1.7 FROM BIDIR BIDIR T T O O R R R R C C G G P P T T N N L F L F I I A T A T U U C C O O O O R R L L E E R R Why hubbed/star traffic pattern? Failure scenarios: • European backbone: gateway to USA • single link (left) OR node (right) failures • --> link failures always interpreted as link failures • Residential ISPs Traffic: • pattern: • Traffic to/from a server farm • uniform: “bidir” • hubbed: “from” or “to” single node • Etc. • Randomly generated (integer values) Topologies • Large: 57 links and 44 nodes
Results: Electrical MPLS Recovery SINGLE (MPLS Rerouting) versus MULTI (OSPF) path for VARYING LINK WEIGHT 1.03 MULTI-/SINGLE-path Survivability COST: 1.02 1.01 Ratio of 1 LINE failures 0.99 NODE failures 0.98 0.97 0.96 1 2 3 4 5 0 5 0 0 0 1 1 2 0 0 1 0 1 MAX LINK WEIGHT Single Path • MPLS Rerouting : single LSP between two nodes, restored by another single LSP Multi Path Failure scenarios: • OSPF : forward packets evenly over all • single link (white) OR node (gray) failures interfaces which have same distance to • --> link failures always interpreted as link failures Traffic: destination • pattern: single, uniform traffic matrix Topologies • MPLS rerouting : consider multiple equal • Large: 57 links and 44 nodes cost LSPs (each to be rerouted!) --> scalability • Link weights: randomly generated problem!
Results: Electrical MPLS • Local Protection > FTCR > End-to-end: – FTCR is a combination of Local Protection and End-to-end • End-to-end: – Rerouting > end-to-end protection or local-loop back: • protection --> less alternative routes --> potentially less spare resources – End-to-end protection = +/- Local loop-back: • downstream no traffic anymore --> place for local loop-back of opposite direction • Hubbed Traffic pattern: – FTCR performs significantly better for traffic from the hub than for traffic to the hub. • Single (MPLS Rerouting) versus multipath (e.g., OSPF) – Working cost identical – Decreasing maximum link weights • Multipath seems to perform slightly better • But also higher variance on multi/single path ratio.
Sharing in MP λ S: local protection Select (= switch to) best signal Select (= switch to) best signal Dedicated, thus 2 Default wavelengths needed Converging backup Tree: AT MOST single output wavelength!!!
Sharing in MP λ S: path protection Independent routing!!!
Sharing in MP λ S: path protection Even if red and black working paths do not overlap, the wavelength cannot be shared on this link, because they are routed differently downstream. • At most 2 working paths through each piece of equipment. Thus at most 2 backup wavelengths needed on each link • Cost backup wavelengths = 10+5sqrt(2) (unit = cost for 1 wavelength per length of horizontal link)
Sharing in MP λ S: path protection • How to force to share backup resources? – Limit routing of backup paths to a predefined/predistributed tree – Why? • Avoid situation that backup paths divert after overlapping • Forcing routing so that as much of the backup route is shared with other routes (even if this results in slightly longer backup routes --> to be compensated by the sharing). • 3 backup routes can share 2 wavelengths • Cost reduced from 10+5sqrt(2) to 12+2sqrt(2)
Sharing in MP λ S: path protection Red and blue should be protected at the same time. To which color has the backup of the black path to be tuned, in order to share the backup wavelengt Conclusion: ingress of black path cannot swap to THE backup OLSP, in combination with simple merging Blue Red downstream.
Recommend
More recommend
