Wednesday, November 7, 2012

Enabling Elastic Optical Networking for SDN Architectures

Software Defined Networking (SDN) is a term that has quickly risen to the forefront of the networking industry, with the objective of solving many of the challenges facing network providers today. While there are many varying opinions of exactly what SDN means, there is one common trait amongst them - “programmability”. 

SDN is a means to open up the network and support programmability of the network, often comprised of multiple vendors, multiple domains, and multiple networking layers. For some providers’ networks, programmability and SDN’s functions are viewed as a means to enabling automated, on-demand networking with optimal resource utilization. 

Initial SDN efforts are largely focused on decoupling the control plane from the data plane and enabling a higher level of programmability into packet forwarding tables of switches. But what does this mean for the emerging converged optical transport layer with integrated switching, which is now playing an increasingly important role in optimizing networks and underlies most of the world’s Internet backbone? What does the optical layer need to adequately and economically support a programmable network with on-demand capabilities?

The Evolution of Optical Transport: Opportunities and Challenges

The architecture of the optical network is undergoing a significant transformation – and with this transformation lays some new challenges around automation, elasticity, and capital/operational efficiency.

The Network Efficiency Challenge

Over the past many years, network providers were deploying 10Gbps wavelengths on a large scale, and the predominant service that was transported over the fiber backbone was 10Gb. Initially, it was SONET/SDH OC192/STM64 service, running at 10Gb rate, and more recently, 10Gb Ethernet (10GbE) services have rapidly risen in popularity, driven by Ethernet convergence. In this environment, the transport service speed matches the wavelength bitrate, and the term “wavelength” was used synonymously to mean the optical, analog transmission wavelength as well as the transparent wavelength-like digital service.

Today, however, there is a growing divergence between the wavelength bitrate and the transport services the network needs to support. Layer 0 transmission is rapidly evolving towards 100Gb optical wavelengths equipped with coherent detection, as carriers and network providers strive to increase fiber capacity to multiple Terabits. But the supported transport services are still largely 10Gb, sometimes less. Moreover, the services mix will continue to include a variety of service rates, as dictated by the economics of these services.

While the quest for achieving optimal cost/bit transmission economics is driving the need for 100Gb wavelength technology, the business of transport services and market demand for a broad set of service rates necessitates a different approach than in the 10Gb wavelength era. With the current economics of 10GbE vs. 100GbE services, it is generally expected that 10GbE services will continue to dominate in volume for some time while networks are upgraded with 100Gb optical technology. Further, this divergence of service from optical wavelength speeds will likely continue, as optical technology takes its next step forward to super-channels – wavelengths with bitrates beyond 100Gb. State of the art technology today offers 500Gb super-channels, with 1Tb super-channels soon to follow, as the drive to increase fiber capacity continues.

The Bandwidth Elasticity Challenge
A second challenge facing providers today is the need for on-demand “elastic” bandwidth to efficiently and cost-effectively deliver bits whenever and wherever needed. Evolving traffic patterns driven by cloud network and datacenter communications are driving providers to relook at their network architecture and the relationship between IP and optical transport layers. The conventional practice of over-provisioning the IP layer and running links at low utilization rates, while constraining the optical layer to provide static, “always-on” 100Gb point-to-point capacity, is being scrutinized, as new optical transport solutions with integrated digital switching emerge that can readily flex and adapt to varying and unpredicted bandwidth needs.

While this level of flexibility and adaptability begets greater elasticity, it highlights the more general challenge of multi-layer resource optimization. With resources that can be allocated and repurposed at multiple network layers, network providers require ways to optimally allocate resources to provide the appropriate bandwidth connectivity services that meet the service requirements of applications.

The Network Automation Challenge

In order for the network to provide on-demand bandwidth at Internet speeds, operational processes need to devoid themselves of human intervention. This encompasses not just automation of processes across multiple network layers, from transmission and transport up through IP/MPLS, but also the orchestration of resources between separate domains and amongst multiple vendors. In the optical transport layer, this means enabling rapid delivery of transport bandwidth in a manner that is cost and resource efficient, without burdensome wavelength engineering processes.

All-Optical Networking Dilemma

Conventional all-optical networks based on ROADMs deliver wavelengths on an end-to-end (A-to-Z) path, and constrain the delivery of transport services statically between those two sites. Operating on the principle of photonic switching, ROADMs can only route entire wavelengths and cannot access transport services carried inside the wavelength. Capacity that is not utilized within the wavelength cannot be leveraged by other traffic demands that do not originate at the same locations but which might share the same physical sub-path. 
As networks evolve to 100Gb wavelengths and beyond, this not only presents a resource utilization challenge, but also a bandwidth elasticity challenge. 

All-optical networks empowered with ROADMs are capable of “flexing” to dynamic bandwidth demands with wavelength granularity only. They facilitate turn-up of new end-to-end wavelengths, but the inability for all-optical networks to 1) manipulate the services inside the wavelength or to 2) pool capacity together and dynamically allocate bandwidth leads to static, underutilized wavelengths and excessive deployed capital. As the pressure to increase fiber capacity grows, leading to larger but fewer super-channels, the wavelength fragmentation challenge is exacerbated.

With the evolution of optical wavelengths from 100Gb today towards 1Tb super-channels in the future, the role of ROADMs will inevitably evolve towards steering large chunks of capacity between major hubs, and less for turning up and delivering digital services to end users. Tightly coupling the allocation of dedicated wavelengths between A-to-Z network locations for on-demand delivery of services does not scale.

Nevertheless, programmability of the ROADM, as well as key optical transmission parameters such as modulation scheme for trading off reach versus capacity, are important elements of SDN for the overall optical transport layer.

In order for a network to offer truly elastic bandwidth, and enable transport bandwidth service to be efficiently delivered over any optical wavelength, virtualization of the wavelengths is a necessity. This entails creation of an abstraction layer that represents the creation of a pool of optical resources that can be leveraged for any bandwidth demands.

The Solution: Bandwidth Virtualization

Facilitating SDN’s programmable networking concept in a manner that simultaneously optimizes utilization of optical capacity and enabling real-time delivery of optimally-sized bandwidth requires a means to decouple transport service delivery from the transmission layer. It requires an abstraction layer that virtualizes wavelengths and pools the capacity together on each link, and promotes sharing of that bandwidth for any transport circuit traversing that link. Instead of a dedicated resource between 2 fixed locations, wavelengths can be transformed into a shared resource supporting services between any network locations.

Not surprisingly, this concept is very similar to IT resource virtualization, where the collective power of multiple physical resources are pooled together and shared amongst multiple Virtual Machines (VMs). VMs supporting applications can be dynamically instantiated or decommissioned from this shared pool of resources, maximizing utilization and efficiency. Bandwidth Virtualization achieves a similar objective by forming an abstraction layer representing a bandwidth pool and hiding details of the underlying optical wavelength resources. Any bandwidth service can be flexibly mapped to any physical wavelength resource on each digital network link, whether the wavelength bitrate is 10Gb, 100Gb, or 1Tb.

This is essential as optical transport evolves towards super-channels. Additionally, through finer granularity switching of transport services rather than coarse wavelengths, Bandwidth Virtualization provides the foundation for SDN programmability. With Bandwidth Virtualization, the transport service provisioning process is decoupled from wavelength engineering, leading to significant benefits including:

  • Reduced time to delivery of new bandwidth services to meet unexpected demands
  • Responsiveness and adaptability of the optical transport layer to dynamic needs of the application and IP layers with appropriately sized optical transport capacity
  • Efficient on-demand allocation of bandwidth from available resources to maximize wavelength utilization
The capability of digitally switching individual transport services rather than just optically redirecting coarse wavelengths provides a level of bandwidth service flexibility that is decoupled from the evolution of optical transmission technology.

Enabling Bandwidth Virtualization

The virtualization of the optical wavelengths requires abstraction into a shared pool of digital bits that can then be rapidly allocated to support any transport service. Key enablers of the Bandwidth Virtualization paradigm include (a) cost-effective OEO conversion to gain accessibility to the individual services being transported via optical carriers and (b) integrated digital switching. Conversion of wavelengths into the electrical domain normalizes the traffic into a form where it can be managed, independent of wavelength bitrate and origin, and enables individual bit-based services to be sorted (demultiplexed), switched, and groomed, before being remapped on to an outbound optical carrier. These functions provide important network capabilities including:

  • Redirection of individual transport services for optimal latency routes
  • Optional redirection for protection against network failures
  • Maximum level of wavelength utilization through mixing and matching of any services on to any wavelengths
  • Mitigation of wavelength reach limitations or wavelength blocking situations
  • Decoupling of the service provisioning process from complex analog wavelength engineering and turn-up
Critical design requirements for Bandwidth Virtualization solutions include minimal latency, minimal space/power, and scalability commensurate with the optical domain. Additionally, Bandwidth Virtualization must be economically viable – implementations that necessitate excessive “boxes” linked together with optics may technically deliver the same capability, but are not as economic as converged solutions with internal integrated switching.

Along with integrated switching, considerations for supporting the broad set of transport services must be made. Generalized optical transport infrastructures typically need to support multiple rates and protocols, and full transparency of the service with stringent performance requirements is mandatory. Additionally, dynamic scaling of the transport service upwards and downwards is also important for optimizing consumption of optical capacity. With this set of capabilities, networks can capitalize on a new level of transport elasticity that provides appropriately sized bandwidth services whenever and wherever needed in the network.

A New Approach: Elastic Optical Transport

The emergence of optical transport with integrated switching is changing the way architects design cloud networks. Instead of the traditional model of “dumb pipes” interconnecting large routers and relegating all bandwidth management functions within routers, network providers now have the option of deploying cost-efficient, flexible optical transport networks with integrated switching and offloading transport bandwidth from routers.

Evolving traffic patterns in clouds coupled with large amounts of data traffic between data centers warrants traffic more optimally being transported and switched within the optical layer, not solely at the more expensive router layer. This evolution towards a more flexible architecture with multiple dynamic switching layers calls for more intelligence in managing not just multi-layer networks, but also networks involving multiple network domains and multiple vendors. The convergence of WDM, OTN and packet bandwidth management functions into the next generation optical transport layer is creating new opportunities for network providers to further reduce the total cost of ownership of their network infrastructure, inclusive of the IP/MPLS layer, while also providing a more scalable, adaptable, and cost-efficient solution that meets the dynamic demands of emerging cloud architectures.

SDN: Ready for Elastic Optical Transport?

The SDN philosophy of decoupling the control plane from the data plane is an important paradigm many in the industry are investigating as a means for automating processes across a multi-layer, multi-vendor, multi-domain network, and orchestrating the many moving parts through network Application Programming Interfaces (APIs) to provide an optimal bandwidth solution for applications that takes maximum advantage of what each network layer has to offer. Centralization of information creates the opportunity to make better over-arching decisions, as it provides globalized visibility across layers, domains, and vendors that is necessary to understand and make appropriate tradeoffs between cost, performance, survivability, and other key SLA metrics.

In order for SDN to be truly useful in multi-domain & multi-layer networks, SDN needs to incorporate not just broader network management functionality, such as network discovery and monitoring and correlation, but also deepen its control to include the emerging next-generation optical transport layer, where integrated switching adds substantial network value and has significant impact on overall network architecture, including what happens at higher layers. This broader vision ensures the entire network stack becomes open and programmable. With expansion of SDN to include elastic optical transport and abstractions like Bandwidth Virtualization, network providers will be able to unlock the real potential of the multi-layer network and fully leverage the resources available at all layers.