University of Essex Department of Electronic Systems Engineering LR

University of Essex Department of Electronic Systems Engineering LR-PON Protection Routing and fast protection in networks of long-reach PONs David Hunter, University of Essex Tim Gilfedder, BT Invited paper AccessNets, September 2006, Athens David Hunter slide 1 Long-Reach PONs Longer backhaul than GPON Edge router ONU ONU < 100 km OLT OLT To other customers Rationale is cost reduction ONU ONU Existing rings and protected ONU ONU chains between local exchange and core become redundant But still need fast protection 50 ms for LR-PON failure, slower otherwise David Hunter slide 2 David Hunter Page 1 University of Essex Department of Electronic Systems Engineering LR-PON Protection LR-PONs in network with single homing Edge Router Each customer accesses one edge router – single homing GPON designed for single homing LR-PONs use GPON protocol Better resilience required – rings and protected chains not present David Hunter slide 3 LR-PON protection – the problem Fast fault recovery is crucial for customer service LR-PON failure – OLT, backhaul or access – O(50ms) Edge router failure – perhaps fire or flood – slower Other faults dealt with by existing core protocols Redirecting traffic from the customer to a backup edge router is only half the problem All traffic elsewhere must be aware of failure Must be re-addressed and re-routed correctly Must be faster than existing Internet routing protocols LR-PON protection implemented at network edge Independent of core topology and implementation Effectively treat core network as one large switch Initially assume protection of Internet traffic David Hunter slide 4 David Hunter Page 2 University of Essex Department of Electronic Systems Engineering LR-PON Protection Generic dual homing concept David Hunter slide 5 LATTE – traffic re-routing Commercial customers have two IP addresses Deduced from one another either directly or via table LATTE places IP datagrams within tunnels IP-in-IP encapsulation Re-direction to destination via second edge router and OLT Re-ranging triggered by notification packet (residential customers) Using similar principles, could also re-route: MPLS Label Switched Paths via label stacking ATM via VPI substitution E1 via interface to SDH management system Signalling always takes place via IP Regardless of what technology is being re-routed LATTE obtains routing table from FROTH Label and Address Translation via Tables at the Edge David Hunter slide 6 David Hunter Page 3 University of Essex Department of Electronic Systems Engineering LR-PON Protection Example of LATTE routing table Prim prefix 177.67.40.0/22 142.31.16.0/25 152.52.4.0/24 Size 1024 128 256 1st OK? No No Yes 2nd OK? Yes Yes Yes OLT (resid) 0.0.0.0 0.0.0.0 Man? No Yes 133.96.44.21 No source David Hunter slide 7 FROTH – LR-PON reachability discovery Gathers and distributes LR-PON reachability information Multicast between all edge routers, not just neighbours IP used for multicast signalling over existing core network Each FROTH router has global knowledge of LR-PON reachability Implemented in a FROTH router attached to each edge router Simpler signalling and computation than OSPF FROTH maintains a table built from this information Selected information then passed to LATTE New information forwarded to other edge routers Robust against failures in the core Edge router failure takes longer to detect Via timeout of hello messages, rather than triggered update Fast Recovery for OLTs via Transmission of Hellos David Hunter slide 8 David Hunter Page 4 University of Essex Department of Electronic Systems Engineering LR-PON Protection Optimising FROTH for LR-PON protection RIP Only neighbours exchange routing information? Uses triggered updates and regularly timed advertisements? Detects faults via a HELLO protocol? Can offer recovery in under 50 ms? Complex algorithm? Uses flooding for routing information? David Hunter OSPF FROTH – optimisation for LR-PON application Yes No Yes No Yes Yes No – faster propagation via multicast Yes – LR-PON failure uses triggered update Only to detect edge router failures Almost always – primary objective of work No – easy to implement and fast Yes – provides speed and resilience slide 9 Yes Yes No No No No Enhancing scalability with areas David Hunter slide 10 David Hunter Page 5 University of Essex Department of Electronic Systems Engineering LR-PON Protection Using multiple operators with areas David Hunter slide 11 Conclusions This work provides insight into many issues It may represent one possible solution Dual homing provides LR-PON protection Simple semi-analytical model based on experimental data 96% of LR-PON failures recover within 50 ms Edge router failure recovery typically takes 100 to 200 ms This model could be refined and developed further Typically 0.02% of network capacity used for signalling Probably practical, but higher than for existing routing protocols Can potentially re-route most protocols Not just IP, although IP is used for signalling For future investigation: Experimental demonstration planned Evaluate other technologies for signalling, and use of DiffServ Tweak FROTH protocol to avoid need for timeouts David Hunter slide 12 David Hunter Page 6 University of Essex Department of Electronic Systems Engineering LR-PON Protection FROTH status packets Contain records with routing information Status packet advertisements sent every few tens of ms Report that edge router and FROTH router are alive Timeout when no advertisements are received More advertisements increase signalling traffic but speed recovery Triggered updates sent using status packets Sent immediately if LR-PON fails or becomes unreachable No timeout – fault reported almost immediately Password included for security Sent to destination directly, and via intermediate routers Enhances robustness First status packet generally arrives sooner than single packet Information time-stamped Prevents propagation of out-of-date records Only new information is forwarded by each edge router David Hunter slide 13 Model of recovery time Fixed delay term Diagnose an LR-PON fault (1 ms) Update FROTH routing table (1 ms) Send table to LATTE (1 ms) Re-range OLT (30 ms) Statistical distribution modelling transit time Pareto (long tailed) tail Remainder is shifted Gamma 0 ⎧ ⎪ (44.44(t − 7.78))3.416 ⎪ f (t ) = ⎨ e 44.44 (t −7.78 ) 0.244 t 1.637 ⎪ ⎪ 0 ⎩ David Hunter t < 7.78 7.78 < t < 8.13 8.13 < t < 300 t > 300 slide 14 From measurements by Corlett, Pullin and Sargood David Hunter Page 7 University of Essex Department of Electronic Systems Engineering LR-PON Protection Summary of modelling results Failure type LR-PON LR-PON Metro Metro Metro No false timeout David Hunter Comm/ Timeout Delay Res interval target T Comm Res Both Both both N/A N/A N/A 50 ms 50 ms 18 ms 18 ms 50 ms 50 ms 100 ms 150 ms 50 ms N/A Prob(t