IPTV Transport with
P2MP MPLS-TE
Jean-Louis Le Roux, Orange France Telecom Group jeanlouis.leroux@orange-ftgroup.com
MPLS World Congress 2007
Outline
IIP TV service growth and requirements IP2MP MPLS-TE Principles IShort Term Deployment scenario with P2MP MPLS-TE
Various approaches for resiliency IAdvanced issues
Computation of Minimum Cost Trees
Fast Reroute with P2MP Bypass tunnels
Integration with NGEN MVPN for Dynamicity and Admission Control
IClosing remarks
Orange IPTV Service Growth
(French Market)
ITotal Bandwidth
1.8G Traffic at the end 2008
z470 LD channels (3Mbps)
z30 HD channels (12Mbps)
ICustomers
> 1 million in Q2 2007
IPTV Transport
Backhaul
TV Servers Core
TV Head-End router
Core Router (CR)
Backhaul/Aggregation Router (BR)
DSLAM IIPTV transport used to rely on ATM P2MP PVC INow SP are migrating to IP/MPLS
Drivers = Need for higher bandwidth, no longer investment on ATM backbones
ITypical figures: 50 PoPs, 500 Backhaul Routers, 10 000 DSLAMs, 41Million TV customers, 500 TV channels, 1.2 Gbps TV traffic
Requirements
IResources optimization: Need for minimum cost trees
Significant bandwidth saving in meshed networks IResiliency: Need to minimize the impact on images upon link or node
failure (sub-50ms failover) IDynamicity and Admission Control
Dynamicity and Admission control allow significant bandwidth savings and CAPEX reduction
zA channel is transported only if there is a receiver zA requested channel is transported only if there are resources
Useful in the Backhaul. Examples: Less than 80% channels requested in primary rings, less than 50% in secondary rings and on DSLAM links
Useless in the core (near 100% channel requested) IOAM: Need for fast tree fault detection/isolation, tree tracing IP2MP MPLS-TE fits in well with these requirements
P2MP MPLS-TE Overview
IP2MP MPLS-TE is an extension to MPLS-TE for setting up explicitly routed MPLS Trees
IIt relies on P2MP extensions to RSVP-TE
See RFC4461 & draft-ietf-mpls-rsvp-te-p2mp  (stable specification) IAllows for Multicast Traffic Engineering
Significant bandwidth savings thanks to Minimum Cost Trees
Admission control and hence CAPEX reduction thanks to bandwidth reservation
Sub-50ms recovery upon link or node failure thanks to P2MP Fast Reroute IOffer strong OAM features (under maturation in the IETF)
P2MP LSP Ping/Trace, P2MP BFD IReally well suited to TV broadcasting
P2MP MPLS-TE Theory of operations
A P2MP TE-LSP from R1 to {R3, R4, R5}
1: P2MP TE-LSP configuration on R1
Destinations: R3, R4, R5 TE parameters: Bandwidth = 2Gbps Tree shape: Minimum Cost IP 232
R3
Config
2: P2MP TE-LSP Computation on R1
32 IP 232
Taking into account TE constraints \"Multicast CSPF\"
IP 232
R1-R2-R4; R4-R3; R4-R5
17 IP 232 28 IP 232 R4
IP 232
R1 R2
3: P2MP TE-LSP Setup initiated by R1 Static route 232/8 -> LSP1
35 IP 232
P2MP RSVP-TE along the computed paths
4: Static Mcast route within the LSP
IP 232
R5
5: Traffic Forwarding
MPLS Replication
P2MP MPLS-TE Theory of operations
A P2MP TE-LSP from R1 to {R3, R4, R5}
1: P2MP TE-LSP configuration on R1
Destinations: R3, R4, R5
TE parameters: Bandwidth = 2Gbps
Tree shape: Minimum Cost IP 232 R3
Config
2: P2MP TE-LSP Computation on R1 32 IP 232
Current Implementations:
Taking into account TE constraints \"Multicast CSPF\" -Dynamic tree computation
IP 232
R1-R2-R4; R4-R3; R4-R5 17 IP 232 28 IP 232 R4
IP 232
#NAME?
R1 R2
3: P2MP TE-LSP Setup initiated by R1 Static route 232/8 -> LSP1
#NAME?
35 IP 232
P2MP RSVP-TE along the computed paths
.A pure broadcast structure
4: Static Mcast route within the LSP
IP 232
R5
5: Traffic Forwarding
MPLS Replication
P2MP MPLS-TE vs IP Multicast (PIM)
I P2MP MPLS-TE: Current Lack of dynamicity that allows its usage in the core only I PIM: No advanced TE, but dynamicity I Relevant design at short term: Combine P2MP MPLS-TE in the core and PIM in
the backhaul
Resiliency
IRedundant TV Head-Ends: Traffic Transmitted twice with same source address, on two distinct Root Core Routers
IRedundant Core: Odd and Even networks
ICore Resiliency ensured by LSP rerouting, Fast Reroute or Backhaul PIM rerouting
ISeveral Approaches for Edge (Root and Leaf) redundancy
Option 1: Two Trees serving distinct leaves and simultaneously active
Option 2: Two Trees serving all leaves and not simultaneously active
Option 3: Two Trees serving all leaves and simultaneously active IA lot of options not covered here INote: Future NGEN Multicast VPN implementation will natively
support Root and Leaf redundancy (not covered here)
Resilency: Analysis
Fast Recovery  Optimality  Required features  
Option 1  -IGP/BPG & PIM convergence  + Traffic on odd and even in nominal case  OK  
Option 2  + + (core) + (root) FRR in the core Root Switchover  + + Traffic only on odd in nominal case  Root Switchover control feature  
Option 3  + Depends on the BFD interval and RPF update  + Traffic on odd and even in nominal case  P2MP BFD and LSP based RPF  
About P2MP Path Computation
P2MP Path Computation Function
ITree computation relies upon available resources and Tree constraints
Optionally, additional constraints for backward compatibility such as \"Avoid nodes that do not support P2MP MPLS\" zP2MP MPLS capability info in the IGP, see draft-ietf-ccamp-te-node-cap
IOffline Tree Computation: Manually or thanks to a TE tool => the path is explicitly configured on the Root LSR
Allows achieving better optimization, but no reactivity to traffic or topology changes
IOnline Tree Computation: Directly on the Root LSR or Remotely thanks to a PCE (see draft-yasukawa-pce-p2mp-req)
Less optimal but robustness and reactivity to traffic or topology changes
IVarious tree shapes: Minimum Cost Tree (MCT), Shortest Path Tree (SPT)…
SPT versus MCT
Tree S1 -> { L1, L2, L3}
S1
Shortest Path Tree Minimum Cost Tree Tree Cost = 7 Tree Cost = 5 => bw saving Max ¨Path Cost = 3
Max Path Cost = 4
I MCT (aka Steiner Tree) provide significant bandwidth savings in meshed topologies
This is not true in star or ring topologies IMCT Computation is NP-Hard => Need for heuristics
Takahashi MCT Heuristic S1
L1 L2
I Leaves are added one by one in the tree, incremental approach I A Leaf is branched on the closest node already in the tree
A SPF to determine the closest node already in the tree, which is then used to branch to the leaf I Quite Good complexity: K*N*log(N) ( K= #leaves, N= #nodes) I Very well suited to grafting pruning, natively supported. Cost = N*log(N)
Grafting requires a simple SPF, no impact on other leaves
Takahashi MCT Heuristic Evaluation
I Simulation in the Orange IP/MPLS Backbone in the US
Average node degree = 4 I For 20 leaves there is more than 60% bandwidth saving compared to SPT I Computation Time depends linearly of the number of leaves I Simple and efficient heuristic that could easily be implemented in routers
About P2MP Fast Reroute
Integration with NGEN MVPN
Dynamicity and Admission Control
Closing Remarks
IIPTV services rapidly growing IP2MP MPLS-TE well suited to IPTV IStable IETF spec, but still only a few implementations INo dynamicity => limit the usage today to the core only INeed for advanced router features
Online MCT computation (Takahashi Heuristic?)
Fast Reroute with P2MP Bypass LSPs IIntegration with NGEN Multicast VPN will bring useful features
Dynamic traffic mapping, Dynamic leaf addition/removal, Admission Control Root/Leaf resiliency…
Allows extending the scope of P2MP MPLS-TE to backhaul networks ITo be deployed in several Orange Group Backbones by end 2007
P2MP MPLS in the core, PIM in the backhaul
z Successful lab testing completed, ongoing field trial
Mid Term target: Integration with NGEN Multicast VPN
References
I Aggrawal, R., Papadimitriou, D., Yasukawa, S.\"Extensions to RSVP-TE for Point-to-Multipoint TE LSPs\" <draft-ietf-mpls-rsvp-te-p2mp>
I Vasseur, J.P., Le Roux, J.L., et al. \"IGP Routing Protocol Extensions for Discovery of Traffic Engineering Node Capabilities\", <draft-ietf-ccamp-te-node-cap-04>
I Le Roux, J.L., Aggarwal, R., \"P2MP MPLS-TE Fast Reroute with P2MP Bypass Tunnels\", <draft-leroux-mpls-p2mp-te-bypass>
I Swallow, G., Nadeau, T., Aggarwal, R., \"Connectivity Verification for Multicast Label Switched Paths\", <draft-swallow-mpls-mcast-cv>
I Rosen, E., Aggarwal, R., et al. \"Multicast in MPLS/BGP IP VPNs\" <draft-ietf-l3vpn­2547bis-mcast>
I R. Aggarwal, E. Rosen, T. Morin, Y. Rekhter, C. Codeboniya, \"BGP Encodings and Procedures for Multicast in MPLS/BGP IP VPNs\" <draft-ietf-l3vpn-2547bis-mcast-bgp>
I Yasuakwa, S., Farrel, A., \"PCC-PCE Communication Requirements for P2MP MPLS­TE\", <draft-yasukawa-pce-p2mp-req>
I Takahashi, \"An approximate solution for Steiner tree problem in graphs\", Math Japonica 24, No 6 (1980),Pages 573-577
THANKS !
MPLS World Congress 2007