For many network operators, the question of how to most efficiently design their core IP/optical backbone network and provide essential bandwidth connectivity to end customers or applications is critical to the success of their business. Methods that work for one network operator may not necessarily work for others and, similarly, methods that worked in the past may not necessarily be the best solutions moving forward. Several factors can influence the choice of approach, but they generally include network topology, traffic characteristics (patterns, flow sizes, and growth rates), traffic dynamism, and target networking applications.
Current practice generally involves interconnecting routers at the port level and using the optical transport layer to create transparent pipes between adjacent routers, tightly coupling router adjacency to point-to-point wavelengths or wavelength services. This approach relegates all packet bandwidth management (e.g., packet aggregation, statistical multiplexing, QoS control, and protection) to the routers. As a result, these fairly expensive routers end up taking on multiple roles, including simple transport functions for transit traffic, where little or no value is provided.
Common network design optimizations at the router layer include increasing router "meshiness" with additional "express" wavelengths to enable core router bypass and offload transit traffic. However, with the general migration we're now seeing from 10 Gigabit Ethernet (10GbE) to 100GbE router port speeds, justification of these express links becomes difficult. At 100GbE rates, router bypass to avoid transit routers requires that all packet flows from the router port be sent to a single next-hop router, which may not be cost-optimal. This is because both the router and transport layers' capacities need to increase simultaneously to support additional interfaces between the optical transport system and the next-hop router.
Another issue with this approach is that any function pertaining to packet processing is performed by the router alone and not by any supporting network layer, so packet traffic must go between network layers to be processed. The IP topology is rigidly tied to and defined by the wavelength services interconnecting the routers. Using an airport analogy, if one imagines the optical network fabric as a big airport hub and the gates as router ports, it's as if all the passengers disembarking at one gate are forced to go to the same next gate. Static point-to-point transparent wavelengths provide limited router interconnectivity options.
A better IP/optical network strategy
Well, what if the router topology and links could be determined by the actual amount of traffic required between endpoints, rather than be constrained by the router's port granularity? What if router connectivity could be decoupled and, instead, be delivered more flexibly by the transport layer in a way that enables network operators to use their 100GbE router ports more efficiently, while also still enabling router bypass to transport bits across the core more efficiently? Is there a way that packet flows from router ports could be transported differently across the backbone with the right amount of optical capacity, based on specific requirements for each different class of service, without having to transit additional routers?
Network operators today are actively exploring a new approach to building the packet-aware optical transport network. This converged optical transport approach integrates high-speed Layer 2/Layer 2.5 packet intelligence and QoS at the edge of or within the Layer 0/Layer 1 optical backbone, infusing the transport network with new flexible packet-level capabilities and giving network operators the ability to perform packet functions natively at the transport layer rather than in the upper router layer.
This marriage of packet and circuit technology isn't a new concept. However, the convergence of Ethernet and MPLS intelligence at 100GbE rates along with appropriately sized dedicated transport bandwidth circuits has only recently been introduced. This ability creates much more flexible transport that service providers can leverage for increasing overall network efficiency at both the IP and optical layers.
The technology integration across Layer 2 and Layer 1 enables a service provider's optical transport network to now provide on-demand, flexible, QoS-differentiated Layer 2 transport services natively for all packet flows from the routers, without having to leave the optical layer. This includes not just point-to-point services but also point-to-multipoint and multipoint-to-multipoint packet transport services.
The benefits of packet-aware transport
The packet-aware approach provides several benefits:
- Enhanced use of 100GbE router ports. By enabling each port to interconnect with multiple packet-aware express lanes, each providing connectivity to another router, the transport layer can now perform packet aggregation functions with QoS control. Routers no longer have to perform this basic function. In our airport hub analogy, this means that passengers disembarking at a gate can now disperse to multiple gates rather than be forced to a single gate.
- Flow-based differentiated bandwidth management. With packet awareness integrated into the transport layer, flows from a single router interface can be service and transport differentiated. This enables them to be switched over different physical transport paths with varying bandwidth profiles and packet QoS attributes, even when interconnecting the same pair of routers. By using various packet-level parameters for differentiating flows, such as priority level, operators can gain finer control over how the flows are transported through their infrastructure. Revisiting the airport hub analogy again, this means each group of disembarking passengers at a gate can take different modes of transportation to get to their next gates, depending on factors such as layover time. Differentiation can also take place at Layer 1.
- Increased transport bandwidth efficiency. With packet-aware transport, the amount of Layer 0/Layer 1 transport bandwidth can be fine-tuned in 1.25-Gbps increments to better match the actual packet traffic that is mapped into the tunnel. This ability provides more granular control over bandwidth resources and reduces stranded bandwidth compared to what is available when using only 100GbE port-granular wavelengths as in the router interconnect approach. This feature reduces or eliminates router transit traffic, and thus reduces both router and transport ports to make the overall network more efficient. Furthermore, these flexible packet-aware tunnels can support hitless resize to accommodate growth or decline in traffic demands. In our airport analogy, this means one can, for example, adapt the size of the inter-gate transport shuttle based on the desired seat occupancy rate.
- Lower network protection costs. The current practice of protecting traffic within routers using MPLS FRR requires over-dimensioning of router links to accommodate failure scenarios. Additionally, a corresponding amount of optical capacity is also required for this IP layer protection bandwidth. With new sub-50-ms priority-based shared mesh protection schemes available in converged optical transport systems, however, this protection can be moved closer to the physical layer where the fiber cut occurs (see "Surviving disasters with fast shared mesh protection"). Protecting packet flows natively within the transport layer reduces the amount of Layer 3 overprovisioning, which means fewer router interfaces. While the transport layer itself needs to be scaled for protection, the net result is increased IP link use, a reduction in router ports, and a net reduction in total network cost.
- Cost-optimized high-speed Carrier Ethernet services. With natively supported high-speed packet functionality integrated into the transport layer, operators can offer more cost-efficient Carrier Ethernet services (E-Line, E-Tree, and E-LAN) directly on the transport systems, without having to incur the cost of transiting the service up to another switch or router and terminating the service there. End customers can connect with a single 100GbE port and gain access to multiple pay-as-you-grow services through the single interface. Operators can control and customize the QoS parameters for these services at both the packet and Optical Transport Network (OTN) layers, providing new differentiated service offerings.
Become aware of packet-aware networks
For service providers, the growth and prevalence of packet-based services that must coexist with still-critical and profitable circuit services are driving the need for networking tools that can efficiently and flexibly handle both types of traffic with optimal economics. Given the tight interdependency between IP and optical, it behooves today's network architects to understand how emerging technologies at each layer might help the whole network. They also must explore opportunities for optimizing traffic, which includes looking at ways to manage specific traffic types differently.
High-speed packet awareness within the transport layer can transform the optical network from simply a provider of point-to-point wavelength services to one that can now provide flexible, intelligent, and differentiated packet transport and aggregation functions. Such a network will lead to improved utilization and efficiency of the combined IP and optical network layers.
The integration of packet-forwarding capabilities into the transport layer also lays the foundation for more advanced Layer 3 capabilities and services that can eventually be supported natively. The evolution towards converged multi-layer transport coupled with a software-defined network (SDN)-centric cloud-based IP control plane, fully decoupled from the physical network, helps paves the way for a new simplified, highly efficient and flexible network model (see, for example, "Transforming Metro Network Economics").
Chris Liou is a Fellow and vice president, network strategy, at Infinera, where he focuses on optical transport networking architecture and multi-layer SDN for content providers, data center network operators and service providers. Chris has also served in the roles of Vice President of Product Planning and Vice President of Product Management at the company.