Cable Metro Packet Optical Transport An Overview and Vision for L0-L7 Convergence A Technical Paper prepared for the Society of Cable Telecommunications Engineers By Andrew Smith Distinguished Engineer & Chief Architect for Cable MSO Networks Juniper Networks 1194 N Mathilda Ave Sunnyvale, CA 94089 +1 408 745 2000 [email protected]
15
Embed
Cable Metro Packet Optical Transport - Juniper Networks · 2014-10-16 · based on MPLS LSP’s to accommodate services as required. The root metric for allocating and assigning LSP’s
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Cable Metro Packet Optical Transport
An Overview and Vision for L0-L7 Convergence
A Technical Paper prepared for the Society of Cable Telecommunications Engineers
By
Andrew Smith
Distinguished Engineer & Chief Architect for Cable MSO Networks
Contemporary cable metro networks have undergone a tremendous transformation in the last
decade. From the early days of IP (Internet Protocol) data as simple Internet access, to modern
networks supporting a multitude of residential and commercial data services, the cable industry has
captured tremendous benefit from packet infrastructures. A high rate of disruptive innovation has enabled
new applications and revenue streams for cable operators, all based upon the high-speed delivery of
packetized, IP data.
This innovation has developed predominately at the packet and application layers. Underneath
this innovation lies an optical transport layer (the “Optical Domain”) which has continued to operate in
isolation of the routing and application layers (the “Packet Domain”). This separation of domains leads to
suboptimal placement of technology and capital, resulting in inefficiencies and persistent operational
challenges.
This paper describes a proposed architecture for a packet oriented, application driven transport
system that enables the next generation of all-packet, all-IP dynamic multimedia residential and
commercial services in modern cable regional networks. In the regional / metro area, due to the relatively
shorter distances involved, unique cable industry use cases, and all-IP based services, we have license
to rethink the status quo network architecture through the adoption of new technology enablers and
disruptive innovations.
Specifically:
- Next generation of optical integration into the packet domain
- Auto-discovery, and automation, of optical attributes
- Ease of packet deployment and transport LSP (Label-Switched Path) establishment
- Multi-layer optimization, including application layer feedback
- Use of modern SDN (Software Defined Networking) derived tools to automate, manage and scale
the infrastructure
Such an evolution is based on the Packet Domain subsuming the Optical Domain so as to enable
a degree of flexibility and programmability and to meet the next generation of SDN enabled services.
Traditional metro network architectures have meant overinvestment in optical transport, or packet
forwarding, or some combination thereof. Through graceful and software driven use of optical bandwidth,
cable operators can reliably transport data across the network while avoiding, to a degree, the very latest
optical technology.
In modern Internet and packet networks operators are accustomed to the many benefits of
statistical multiplexing (stat-mux) of data. Operators routinely “overprovision” the networks. The
fundamental architecture of any-to-any, packet based, connectionless networks permits reliable service
delivery. Stat-mux gain is increases with increasing discrete flows on the network and increasing
variation of each flow. Extending packet stat-mux to the optical layer and controlling the network with
modern software results in a totally optimized packet network .
In Figure 1, a circuit switched network would require a circuit the size of P1 (at a minimum) and
another circuit of size P2 (at a minimum). The granularity is very coarse, as an ODU0 (Optical Data Unit)
is 1.25 Gb/s. In packet switched networks, the required bandwidth in the core is actually less than P1 +
P2.
Figure 1: In OTN or circuit switched networks, circuits must be provisioned to peak capacity between each router. In packet transport systems, this is not necessary. A “PE” is an edge router in a service provider
network.
2) Optical Integration
This paper proposes total integration of the optical transponder and multiplexor into the router
(Packet Domain). In traditional architectures, the transponder and multiplexor reside in an external
element within the Optical Domain. This separation of optical and packet networks imposes discrete
Peak%value%P1%
Bit-rate%
/me%
Peak%value%P2%
Bit-rate%
/me%
PE1%to%PE2%traffic%
PE3%to%PE4%traffic%
OTN%circuit%size%
OTN%circuit%size%
Wasted%Bandwidth%
Wasted%Bandwidth%
PE1% PE2%
PE3% PE4%
Core%
B/W%?%
control and management planes, which require separate engineering efforts and unique disciplines in
network planning.
With the development of modern 100G coherent technology, the transponder can be gracefully
integrated into the Packet Domain, creating an optimal relationship between optical and packet
forwarding capacities, a reduction of operational expenditure, and in some cases, an improvement in
optical performance. No longer is there a gap between the packet forwarding capacity in the packet
domain and the supportable bandwidth in the optical domain. The relationship between the two becomes
a tight bond, with significant optimizations to the operator.
Future optical components are expected to continue their trajectory of miniaturization and power
reduction, bandwidth improvements, and industry efforts such as OpenFEC will permit multi-vendor
interoperability. OpenFEC is an effort to create open, publicly documented Forward Error Correction
standards, with one intent being to create vendor standardization and interoperability between optical
systems. These developments will enable continued phases of core evolution, and will enable the
densities and power envelopes required for the packet domain to subsume the optical domain. This
integration of the multiplexing and demultiplexing of lambdas, or individual wavelengths, into the packet
domain enables an opportunity to rethink how a transport domain supports IP services.
Previous methods to integrate the optical transponder into the packet domain required the use of
a packet-to-optical control plane integration scheme. A suite of protocols known as GMPLS (Generalized
Multi-Protocol Label Switching) was leveraged to permit the router to signal a wavelength through the
optical domain and various “split management” techniques, such as Virtual Transponder, to permit the
optical domain to manage the router’s transponder as if it were still an element in the optical domain.
While these methods of packet optical integration represented a significant optimization of packet
resources and operational savings, it imposed the cost of a complex control and management layer.
These contemporary solutions represent only an incremental evolution towards transforming the network
core, and this paper discards it’s use.
Modern data networks in cable systems are dominated by router-to-router IP services and data
center interconnection. As shown in Figure 2, this results in two disparate “systems” – optical and packet
– that have independent optimization points. This architecture proposes to merge those layers into a
collapsed packet-transport network fabric, resulting in an aligned, congruent and flexible system.
Figure 2: Traditional networks have separate transport and packet layers, resulting in inefficiency. Merging packet and optical results in an optimal architecture.
3) Auto Discovery and Automation
Mul$%layer*Meshed*Network*
Collapsed*Packet%Transport*Network*Fabric*
The packet domain is accustomed to a model where network elements discover one another
dynamically, exchange capabilities, and subsequently build the required databases and forwarding state
to enable the switching of services. Once the optical transponder and multiplexing and demultiplexing
functions are integrated into the packet layer, new opportunities for discovering total multi-layer
connectivity become available. In this evolved architecture, the integrated transport system elements
will dynamically exchange optical characteristics through ‘Optical Peer Discovery’ procedures, automating
the wavelength and channel setup by simply defining how much packet bandwidth is required between
packet nodes.
Autodiscovery of adjacent packet elements will happen through the traditional Optical Supervisory
Channel (OSC), which has long been a part of contemporary DWDM (Dense Wave Division Multiplexing)
optical systems. Elements will exchange wavelength allocation policies and spacing requirements over
the OSC. Existing protocols, such as LLDP (Link Layer Discovery Protocol) and Netconf, can be
leveraged to discover, manage, configure and operate elements along the path. Through the automatic
discovery of optical infrastructure, such as amplifiers, and the coordinated exchange of wavelength
requirements between routing elements, a “plug and play” functionality can be brought to the optical
domain that has long been the standard operating model of the packet domain. As shown in Figure 3,
much how routing protocols discover adjacent nodes and exchange properties, this architecture proposes
a similar mechanism at the optical layer.
Figure 3: Packet Optical devices discover their neighbors, exchange capabilities, and establish service. Routing protocols do this today in the packet domain.
4) Simplified Packet Configuration and Centralized Path Optimization & Management
In traditional packet domain architectures, the network operator must choose specific
configuration items for the routing infrastructure, with a mix of IGP (OSPF, IS-IS), MPLS (LDP, RSVP)
BGP (multiple address families) and related protocols. Originally intended to permit maximum flexibility to
the operator, the sheer number of permutations these choices enable can create an environment of
uncertainly and perceived complexity of what is otherwise a very elegant and graceful packet
architecture.
As a product of the specific use case of this architecture, and by leveraging advances in Software
Defined Networking technologies, the configuration of the MPLS infrastructure to support packet transport
is radically simplified. Simple, and automated configuration of the IGP and the label distribution protocol
Peer$Discovery$
Property$Exchange$
1Tb$Transport7group$
800G$Transport7group$
is all that is required, actual path computation and LSP establishment is delegated to the Path
Computation Engine (PCE).
A PCE is a critical element of this evolved architecture. The PCE provides the central function of
LSP provisioning and resource allocation within the transport domain. By leveraging the Stateful PCE
extensions of the Path Computation Element Protocol (PCEP), a PCE can compute constrained paths,
provide reliable state synchronization and optimize both existing and new LSP’s.
Leveraging the power of MPLS Traffic Engineering (MPLS TE) in packet optical transport
networks enables an elastic, load-adaptable and highly available architecture. MPLS TE permits the
network to respond dynamically to changing traffic conditions, and to adapt the network to the demands
of the modern cable regional core network. The analytics produced by the Path Computation Element is
the origin of the advanced logical topology, instantiated by MPLS TE, and delivered over a packet plus
optical domain. Figure 4 shows the work flow of various protocols and software processes. The concept
is to retrieve data from the network and applications, compute an optimized path for all demands on the
network, the install that path onto the network. Discovery of network information can come from PECP,
from BGP Linkstate (BGP-LS) or other external sources, instansiation of the topology to the network can
be deployed via PCEP, BGP, OpenFlow, I2RS or other data plane control protocols.
Figure 4: The power of the Path Computation Element, Software Defined Networking and MPLS Traffic Engineering creates an analytically driven optimization system for packet optical transport networks.
5) Multi-layer optimization, with application feedback
The Path Computation Engine, by virtue of its role in optimizing services on the network, has a
multi-layered view of the transport infrastructure:
- The IGP (ISIS or OSPF) and the optical infrastructure (Layer Zero) are congruent, meaning there is
no ambiguity as to the physical path that traffic will take;
- MPLS LSP’s, pushed into the network via PCEP, are congruent with specific service demands and
requirements of the network, so there is no ambiguity as to the logical path traffic will take.
The contemporary use case of PCE technology is to reconcile the demands, constraints and
requirements of layers zero through three, and produce an elegant logical packet forwarding topology