PLNOG 6: Maciej Konstantynowicz - Implementing Seamless MPLS

Implementing Seamless MPLS
Maciek Konstantynowicz, maciek@juniper.net
Internet Engineering, Juniper Networks
March 2011

2 Copyright © 2011 Juniper Networks, Inc. www.juniper.net
Juniper Confidential
SP Priorities
  Minimize recurring operational costs
!  Simplify provisioning, management and troubleshooting
!  In many cases IT Systems struggling with keeping up with next-gen network
evolution
  Improve service velocity
!  Reduce time to service, lower new service intro cost
!  Improve effectiveness of service delivery
!  Differentiate service offering in the market place
  No compromise on service quality
!  E2E restoration with transparency of network failures to service layer
!  Appropriate QoS SLAs and quality of experience per service type
  ! Must factor in bandwidth and service mix uncertainties, not an
easy task

More about service trends..
  Most edge services today are delivered at the boundary of metro
and backbone – in the IP POP
!  Some services are even more centralized incl. mobile and Data Center
  However service dynamics are changing rapidly due to following factors
!  Compute power of end devices, incl. increased number of mobile devices/tablets
!  Increasing network access speeds with fiber access
!  High volume video content, both walled garden and over-the-top
  De-centralization of selected services is becoming reality
!  Residential services, location-based services, video caching
!  But some services likely to stay more centralized (e.g. business L3VPN)
Target packet network architecture must enable this seamlessly

Translating into Requirements for
Converged Packet Network
  A single converged packet network for delivery of all services
!  Wireline, mobile, residential, business, wholesale
  Support for large network and services scale
!  A single packet network spanning core, edge, aggregation and access
  Deterministic availability SLAs regardless of scale
!  Fast service restoration and transparency of network failures to service
layer are key drivers for network consolidation
!  Enables migration away from SONET/SDH
  Greatly simplified service introduction, delivery and operations
!  Flexible topological placement of service delivery points
!  Minimized number of service provisioning points

A Word on packets vs. circuits
  Common perception
!  MPLS ports expensive, OTN ports cheap
  Reality check
!  Depends on service demands and topology, quantifying statistical multiplexing gain is key
  Need network cost modeling for multi-layer optimization
!  New modeling approach enables cost optimization based on realistic packet flow approximations(1)(2)
!  Packet demands modeled more accurately taking into account avg BW, peak BW and concurrent peaks
!  Cost optimization problem formulated as a linear optimization using MPLS and OTN ports and devices
!  Result: a provable optimal solution, i.e., there is a guarantee that no better solution exists
References:
(1) “MPLS vs. OTN”, K.Kompella, Juniper, MPLS & Ethernet WC Paris, February 2011.
(2) “Robust optimization models for networks with statistical multiplexing”, P.Belotti et al, Clemson University, September 2010.
A new model for packet traffic:
!  Add averages, but consider some
number of simultaneous peaks
!  Describe packet demands as
function of three parameters:
B = average bandwidth
" = peak:average ratio
K = # of simultaneous peaks

MPLS Overview: Why is MPLS useful at all ?
  Separation between Control Plane and Data plane
  Unified Data plane
!  “Magic Carpet” for Services
  Support for arbitrary Hierarchy
!  Stack of MPLS labels
!  Used for Services, Scaling and fast service Restoration
R1 R2 R3
RX 100K IPv4 routes TX 100K IPv4 routes
inet.0
100K IPv4
FIB entries
R1 R2 R3
RX 100K Labeled routes TX 100K Labeled routes
mpls.0
1 MPLS
FIB entry

!"#$%&'()*+,"(+-.(/-.0#%0/-.(1"&23*"(&"1#%&+4%-(
!  !""#"""$%&'%()*+,)$%+-%./0%12,3)4%-)45&63%
!  7)289)$$%$)6*+,)%6),&*)6:%'6&8%2-:%'2+9;6)%)*)-4%<7;=>?"8$@%
5"*%67,".(-"#$%&'(+-.(1"&23*"(+&*83#"*#6&"1(
!  A&819)4)%*+64;29+B2C&-%&'%-)45&63%$)6*+,)$%
!  D9)E+=9)%4&1&9&F+,29%192,)8)-4%&'%$)6*+,)$%G%)-2=9)6%'&6%1)6%$)6*+,)%()>,)-4629+B2C&-%
!  H+-+8+B)(%-;8=)6%&'%16&*+$+&-+-F%1&+-4$#%$+819+I)(%)-(>4&>)-(%&1)62C&-%
Implementation: Seamless MPLS
Foundation for One Converged network
Networking at scale without boundaries
Edge CoreAccessClients Data CenterMetro Aggregation
Seamless MPLS

Juniper’s Seamless MPLS
Functional Blueprint
  Devices and their roles
!  Access Nodes – terminate local loop from subscribers (e.g. DSLAM, MSAN)
!  Transport Nodes – packet transport within the region (e.g. Metro LSR, Core LSR)
!  Border Nodes – enable inter-region packet transport (e.g. ABR, ASBR)
!  Service Nodes – service delivery points, with flexible topological placement (e.g.BNG, IPVPN PE)
!  Service Helpers – service enablement or control plane scale points (e.g. Radius, BGP RR)
!  End Nodes – represent customer network, located outside of service provider network
  Regions
!  A single network divided into regions: multiple Metro regions (leafs) interconnected by WAN backbone (core)
!  Regions can be of different types: (i) IGP area, (ii) IGP instance, (iii) BGP AS
!  All spanned by a single MPLS network, with any to any MPLS connectivity blueprints (AN to SN, SN to SN, AN
to AN, etc)
  Decoupled architectures
!  Services architecture – defines where & how the services are delivered, incl. interaction between SNs and SHs
!  Network architecture – provides underlying connectivity for services
Metro-2 RegionWAN Backbone RegionMetro-1 Region
TN TN BN TN TN BN TN TN AN ENANEN
SH SH
SNSN
Seamless MPLS Network

Network service provisioning
and operation points:
Connectivity – provisioned
by NMS or AAA
L3/L3+ Services –
provisioned by NMS or AAA
Internet
Metro-2 RegionWAN Backbone RegionMetro-1 Region
TN TN BN TN TN BN TN TN AN ENANEN
SH SH
SNSN
Seamless MPLS Network
Juniper’s Seamless MPLS Architecture
Connectivity and Services Blueprint
EN EN
EN EN
EN
EN
SN SN
SN
SN
Basic Pt-to-Pt Connectivity Services
L3 or L2 VPN Services
Content / hosted app. Services
Internet Access Services
Centralized
Business edge
Centralized
Business edge
De-centralized
residential edge
De-centralized
residential edge
Pseudowire
Pseudowire
Pseudowire
Pseudowire
Pseudowire
Any2Any
Any2Any
Any2Any
AN
ANAN
AN
AN
AN
S C
SN
SN
S
CC
C
C
C
S
S
S
C

Seamless MPLS
Simplified Service Delivery
CPE
AN
PE PE
CPE
AN
Aggregation AggregationCore
PE PETraditional
Aggregation AggregationCore
Simplified Service Instantiation
(single provisioning point per connection)
Seamless
MPLS 1 MPLS 2 MPLS 3
One Converged Seamless MPLS Network
Services
AN
BN BN
AN
EN EN
SN SN

Juniper Seamless MPLS
Service and Network Architecture
  Requirements addressed across the three main architectural dimensions
  (1) Scale – enables 100,000s of devices in ONE PSN network
!  Large network scale via MPLS LSP hierarchy and robust network protocol stack (IGP, BGP)
!  No service dependency whatsoever – all packet services supported
!  Low-cost/low-end access devices accommodated natively without adding complexity (MPLS
labels on demand)
  (2) E2E service restoration – enables sub-50ms recovery from any event
!  Service restoration made independent of scale, services and failure types
!  Achieved with full coverage of local-repair mechanisms for sub-50ms restoration
!  Deterministic for any failure domain size / radius
  (3) Decoupled network and service architectures
!  Flexible topological placement of services enabled via MPLS Pseudowire Termination into
Services
!  E2E virtualization of network service delivery with tight integration of Ethernet, IP and MPLS
!  Minimized number of provisioning points, simplifying service delivery and IT systems(!)

Seamless MPLS – EU Use Case
Network Scale
  Design
!  Split the network into regions: access, metro/aggregation, edge, core
!  Single IGP with areas per metro/edge and core regions
!  Hierarchical LSPs to enable e2e LSP signaling across all regions
!  IGP + LDP for intra-domain transport LSP signaling
!  RSVP-TE as alternative
!  BGP labeled unicast for cross-domain hierarchical LSP signaling
!  LDP Downstream-on-Demand for LSP signaling to/from access devices
!  Static routing on access devices
  Properties
!  Large scale achieved with hierarchical design
!  BGP labeled unicast enables any-to-any connectivity between >100k devices – no
service dependencies (e.g. no need for PW stitching for base VPWS service)
!  A simple MPLS stack on access devices (static routes, LDP DoD)

CPE CPEAGN1 AGN1AGN2 AGN2
ABR
RR3107
ABR
RR3107LSRLSR
BGP-LU BGP-LU
ISIS-L1 + LDP-DU ISIS-L2 + LDP-DU ISIS-L1 + LDP-DU
Static-Route +
LDP-DoD
Static-Route +
LDP-DoD
IP/MPLS Network Infrastructure
RR
BGP-LU
RR
ABR ABR
TNTNAN BN TN TN BN TN TN AN
Seamless MPLS Roles
EN EN
push PW-L
push LDP-L
PW-L
swap BGP-L
push LDP-L
PW-L
BGP-L
swap LDP-L
PW-L
BGP-L
swap LDP-L
PW-L
BGP-L
swap LDP-L
PW-L
BGP-L
pop LDP-L
PW-L
swap BGP-L
push LDP-L
PW-L
BGP-L
pop LDP-L
PW-L
pop BGP-L
pop PW-L
Data flow
Network
Control
Plane
Data
Plane
Service
Control
Plane
Targeted LDP
MPLS data plane
Pseudowire
NHS no NHSNHSno NHS
LDP DoD – LDP Downstream on Demand, RFC5036
LDP DU – LDP Downstream Unsolicited, RFC5036
BGP LU – BGP Label Unicast, RFC3107
NHS – BGP next-hop-self

Scale Enablers
LDP Downstream-on-Demand (LDP DoD)
  IP/MPLS routers implement LDP Downstream Unsolicited (LDP
DU) label distribution
!  Advertising MPLS labels for all routes in their RIB
!  This is very insufficient for Access Nodes
!  Mostly stub nodes, can rely on static routing and need reachability to a small
subset of total routes (labels)
  AN requirement addressed with LDP DoD
!  LDP DoD enables on-request label distribution ensuring that only required
labels are requested, provided and installed
  LDP DoD is described in RFC5036
!  But not widely available in IP/MPLS routers apart from MPLS over ATM/FR
!  This is being fixed now "

AGN1b AGN2a
LDP DoD
LDP DoD – Seamless MPLS Use Case
Configuration and Operation
AGN1b AGN2b
IP/MPLS
Backbone
LDP DU
iBGP LU
Static routes:
! 0/0 default
! /32 destination
Static route:
! /32 AN loopback
IGP (ISIS,OSPF)
DSLAM
OLT
DSLAM
OLT
IP/MPLS Network
ABRa
ABRb
IGP
LDP DU
3
1
2
4
5
7
8
①  AN: provisioned static routes
②  AGN1: provisioned static routes
③  AGN1: statics redistributed into IGP
(optional)
④  AGN1: statics redistributed into BGP-LU
⑤  AN: LDP DoD lbl mapping requests for FECs
associated with /32 static routes and
configured services using /32 routes
matching default route(*)
⑥  AGN1: LDP DoD lbl mapping requests for
static route /32 FECs
⑦  AGN1: AN loopbacks advertised in iBGP LU
⑧  AGN1: if (3) AN loopbacks advertised in LDP
DU 6
(*) Requires LDP support for longest match prefix in RIB (in addition to the exact match) as per RFC5283.
LDP DoD – Label Distribution Protocol, Downstream on Demand distribution, RFC5036
LDP DU – Label Distribution Protocol, Downstream Unsolicited distribution, RFC5036
BGP LU – Border Gateway Protocol, Label Unicast extensions, RFC3107

Scale Enablers
BGP Labeled Unicast (RFC3107)
  BGP-LU enables distribution of /32 router loopback MPLS FECs
!  Used between Seamless MPLS regions for any2any MPLS reachability
!  Enables large scale MPLS network with hierarchical LSPs
  Not all MPLS FECs have to be installed in the data plane
!  Separation of BGP-LU control plane and LFIB
!  Only required MPLS FECs are placed in LFIB
!  E.g. on RR BGP-LU FECs with next-hop-self
!  E.g. FECs requested by LDP-DoD by upstream
!  Enables scalability with minimum impact on data plane resources – use
what you need approach

E2E Service Restoration
  Design
!  IPFRR/LFA for local-repair of transit MPLS link and node failures
!  TE FRR as alternative to LFA
!  LSP tail-end protection for egress PE node failures (IP, L3VPN, L2VPN, BGP-LU, RR-
NHS)
!  Optimized global-repair as fall-back if local-repair not feasible (e.g. no LFA cover)
!  Note: LFA cover can be extended with RSVP-TE
!  BGP PE-CE link local-repair protection for BGP edge link failures (IP, L3VPN, L2VPN,
BGP3107)
  Properties
!  Local-repair for all PE access links, PE and P nodes
!  Local-repair for all PE/P transit links, topology independent (albeit certain topologies may
introduce increased complexity e.g. RSVP-TE if no LFA coverage)
!  E2E restoration in O(50ms) achievable, regardless of network and service scale

End-to-End Restoration
Local vs. Global Repair
link break, local-repair start
local repair stop
global repair stop
20 - 50ms
200 – 1000+ ms
  Local-repair complements Global-repair
!  Local-repair keeps traffic flowing while
!  Global-repair gets things right
!  Variation of “Make before break”
global repair start
  Local-repair
!  Based on the pre-computed local backup
forwarding state - provides sub-50msec
restoration
  Global-repair
!  Requires signaling to take place after failure
detection - can provide sub-1sec or longer
restoration times

End-to-End Restoration
IP/MPLS Local-Repair Coverage – 100% Achieved!
  Ingress: CE-PE link, PE node failure
!  ECMP, LFA
  Transit: PE-P, P-P link, P node failure
!  LFA based on IGP/LDP; if no 100%
LFA coverage, delta with RSVP-TE
!  RSVP-TE FRR
  Egress: PE-CE link failure
!  BGP PE-CE link local protection
  Egress: PE node failure (new)(*)
!  LSP tailend protection with context label
lookup on the backup PE
!  Failure repaired locally by adjacent P
router using LFA (or TE-FRR)
  Packet based networks finally can provide E2E service protection similar to
SDH 1:1 protection, regardless of network size and service scale
  This provides network layer failure transparency to service layers,
becoming a major enabler for network consolidation
(*) “High Availability for 2547 VPN Service”, Y.Rekhter, MPLS&Ethernet World Congress, Paris 2011.

Decoupled Network and Service Architectures
  Design
!  Use MPLS transport pseudowires (PW) to virtualize access for L2 and L3 services
!  Service Node (SN e.g. PE, BNG) to support a PW Headend access interface with all
required data plane and control plane functions (HQoS, security, OAM, PE-CE
routing)
!  Combined SN and TN - enable co-existence of IP/MPLS Service and Transport
functions on the same physical node and the same physical links
  Properties
!  L1/L2 access interface on SN replaced with PW Headend virtual interface
!  Access side SN reachability govern by IP/MPLS – decoupled from L1/L2 interfaces
!  Reduced number of access provisioning points vs. present mode of operation
!  SN support for all L2 and L3 services with appropriate scale
!  E2E service restoration with local-repair for SN failures incl. access PW, node,
transit links

MPLS
CoreAccess / AggregationCPE Edge
IP / IPVPNPWE3 / VPLS
xdsl/Eth
vc/1q
vrf
vrf
inet
Pseudowire
connections
L2 VLAN / VC
connections
Connection
provisioning
MPLS
CoreAccess / AggregationCPE Edge
MPLS EthEth
IP / IPVPNPWE3 1q/1ad1q/1ad
xdsl/Eth
vc/1q
vrf
vrf
inet
Pseudowire Termination into L3 services
  Flexible topological L3 edge placement
!  Virtualized pseudowire access interface enables L3
edge insertion anywhere within the MPLS cloud
!  Services can be placed optimally based on network,
services and operation economics – no impact on the
network architecture
  Simplify network operations
!  End-to-end MPLS across Aggregation and Core
domains
!  No Ethernet L2 interconnect with associated
complexity (provision, assure, protect)
!  No complex L2/L3 redundancy schemes (ie no MC-
LAG)
!  Simplified provisioning
  Unify L3 edge
!  Standardize on pseudowire access with Ethernet and
IP encapsulations only
!  Support legacy access (ATM, FR, SONET/SDH) thru
IP i/working function distributed into access/aggr
!  Results in a uniform L3 Edge for all access

In fu
Pseudowire e2e L2 service (eg SDSL)
Pseudowire access to L3 services
Pseudowire access to BNG
Pseudowire L2 wholesale services
VSI
VPWS with
Pseudowire stitching
VPLS
S
VPN
L2VPN wholesale services
Service Architecture and Connectivity
PW Headend Termination for subscribers
sessions (PPP, IP)
PW Headend Termination into L3 services
(IPv4/v6, 6PE, VPNv4/v6)
Single Segment
pseudowire
PE
PE
PE
TNTNAN TN TN TN TN BN TN TN AN
Seamless MPLS Roles
EN ENBN
SN
BNG
SNSN
CPE CPEAGN1 AGN1AGN2 AGN2
ABR
RR3107
ABR
RR3107LSRLSR
Placement
Flexibility

In Conclusion!
  Seamless MPLS approach addresses all key requirements for
converged packet network design
!  Support for all packet services across fixed, mobile, business,
residential, wholesale
!  Support for large scale incl. high number of low end access
devices
!  E2E fast restoration sub-50msec for all network failures
!  Simplified service delivery with flexible topological placement

References
!  draft-leymann-mpls-seamless-mpls-03, N.Leymann et al, March 2011.
!  “High Availability for 2547 VPN Service”, Y.Rekhter, MPLS&Ethernet WC, Paris
2011.
!  “LSP Tailend Proteciton”, H.Gredler,
!  “Seamless MPLS”, K.Kompella, MPLS WC 2009.
!  “MPLS vs. OTN”, K.Kompella, MPLS & Ethernet WC Paris, February 2011.
!  “Robust optimization models for networks with statistical multiplexing”, P.Belotti
et al, Clemson University, September 2010.

PLNOG 6: Maciej Konstantynowicz - Implementing Seamless MPLS

PLNOG 6: Maciej Konstantynowicz - Implementing Seamless MPLS

More Related Content

What's hot (18)

Similar to PLNOG 6: Maciej Konstantynowicz - Implementing Seamless MPLS (20)

Recently uploaded (20)

PLNOG 6: Maciej Konstantynowicz - Implementing Seamless MPLS