OEO vs. OOO,and the winner is...optical networking
There has been much debate, often heated, in the optical-networking industry over the merits and benefits of the optical (optical-electrical-optical, or OEO) switch versus the purely photonic (optical-optical-optical, or OOO) switch. Numerous proclamations that the photonic switch would soon obsolete the optical switch are simply unfounded if an overall network view rather than network-element (NE) view is taken. The photonic switch and its inherent benefits actually complement the benefits of the optical switch.
The value of a photonic switch is indeed promising albeit limited today, due to the manner in which they are interconnected over optical lines already deployed in the networks of today. Frequent debates often compared the photonic switch to the optical switch strictly at the NE level in isolation of their actual network applications.
The realizable values of optical and photonic switches are highly application- and optical-line-dependent. The optical switch is better suited at present to take full advantage of existing network deployments and provide increased flexibility and scalability for new builds and existing build upgrades.
Numerous technologies are currently jockeying for leadership in the photonic-switch market as the base core-switching technology—micro-electromechanical systems (MEMS), bubble jet, and liquid-crystal technology—with MEMS variants at the forefront of the race. The primary and obvious difference between photonic and optical switches is the core technology chosen for the switch fabric.
Optical switches implement switching fabrics that are electrical in nature and thus dependent on the specific line rate and line format of the actual signals being switched. Photonic switches implement technologies that switch signals purely in the optical domain. The different switch-fabric technologies actually complement rather than obsolete each other since they enable inherently different values (see Table).
The primary touted advantage of a photonic switch is that it effectively removes the necessity for costly OEO conversions, thereby reducing cost—less electronics for reduced overall power consumption and footprint. At first glance, that is indeed quite impressive; however, it cannot be admired in complete isolation from the rest of the optical network. For example, by retaining the switched wavelengths strictly in the optical domain, it is more difficult (and costly) to achieve the following functions:
- Performance monitoring—critical to operations staff for maintaining network health.
- Sub-lambda grooming—necessary for flexible and optimal traffic management.
- Advanced and flexible embedded protection schemes.
- Dynamic line equalization.
Performance monitoring (PM) is an absolutely critical feature for the operations staff of any carrier to proactively maintain ongoing network health and aid in the inevitable troubleshooting activities caused by failed components (e.g., transmitter) and/or accidental errors (e.g., fiber cut). By maintaining all photonic-switch traffic purely in the optical domain, performing PM becomes a more difficult task indeed. Solutions such as tapping off received traffic mandates the implementation of additional OEO circuitry on each and every port serving separate optical lines.
If SONET/SDH PM information is required to ensure service-level agreements, this monitoring circuitry quickly becomes line-rate- and format-dependent. Ironically, when the industry moved from regenerators to optical amplifiers (erbium-doped fiber amplifiers—EDFAs), the same concerns regarding reduced PM information were also raised but eventually subsided over time as mindsets changed and adapted. The optical switch, by virtue of its OEO architecture, is implicitly more conducive to PM. The monitoring of signal overhead of globally ubiquitous SONET/SDH signals is easily and cost-effectively achieved.
A truly flexible switch, optical or photonic, must be able to dynamically groom and switch/route incoming traffic regardless of line rate or line format. Photonic switches have granularity from a single wavelength and to an entire port comprising many wavelengths. But since sub-lambda granularity is not possible in the purely optical domain today, this characteristic mandates that sub-lambda grooming take place at the edges of the core network instead, which is inherently less flexible in most network configurations. Optical switches enable grooming from one STS-1 up to the maximum supported line rate—10 Gbits/sec today and 40 Gbits/sec tomorrow.
The purely optical nature of the photonic switch, coupled with already deployed DWDM line technology, will hinder its wide scale deployment. Although the photonic-switch fabric itself may be nonblocking in nature, the required passive DWDM multiplexing/demultiplexing filters attached to the photonic switch required for DWDM line systems will likely create blocking scenarios.
Numerous methods exist for providing flexible protection schemes. For example, the channel bit-error rate may be monitored at the electrical layer until thresholds are surpassed, thereby enabling protection switches. Alternately, optical layer parameters may also be monitored to provide a mechanism for initiating protection switches.
Advanced protection schemes monitor numerous different electrical and optical parameters before initiating a protection switch and the rerouting of traffic. The minimum granularity that photonic switches can reroute via protection switches is at the wavelength level, meaning that all sub-lambda traffic is forced under the same protection scheme and routing path.
Optical switches can assign different protection schemes to each wavelength and/or sub-wavelength traffic, allowing service providers to offer different classes of service and price them accordingly as a competitive advantage. Optical switches may implement signaling traffic along with the payload traffic or externally using distinct control planes.
The line side of the photonic switch composes a chain of optical amplifiers (Raman and/or EDFA) that must be equalized, such that each received signal exhibits the appropriate optical signal-to-noise-ratio (OSNR) value. That is required to ensure a robust and error-free optical transmission link.
Solutions for network equalization available today range from manual to fully automated, albeit in proprietary manners, since no industry equalization standards are yet available. Proper equalization of any given link is inherently dependent on the following highly interrelated factors:
- Specific wavelengths propagating in a given fiber.
- Number of co-propagating wavelengths in a given fiber.
- Chromatic dispersion profile of the overall end-to-end fiber route.
- Polarization-mode-dependent profile of the overall end-to-end fiber route.
- Attenuation profile of the fiber route.
- Nonlinear effects—four-wave mixing, stimulated Raman scattering, self-phase modulation, etc.
- Others—coding schemes, amplifier characteristics, filter characteristics, etc.
Unfortunately, all of these factors are highly interrelated and co-dependent, making proper equalization a nontrivial task. That is further exacerbated if wavelengths are dynamically added/removed from equalized links carrying live revenue-generating customer traffic. Equalization is highly path-dependent in that the specific fiber type and route distances play significant roles in the characteristic of the link in question. Equalization of these links takes time to stabilize, making the rapid and dynamic routing of wavelengths in an active mesh topology quite involved. A large mesh network comprising numerous wavelengths being switched from link to link yields different equalization scenarios since each switched wavelength results in different combinations and characteristics of the above factors.
Legacy optical line technology already deployed does not allow for the rapid and dynamic equalization of networks in a standardized manner, since it was never intended to do so at the time. By deploying optical switches, the dynamic routing of traffic is facilitated at the electrical layer over existing optimized optical links since the wavelengths already deployed are not rerouted, thereby facilitating mesh networking.
Scalable capacity. The current financial downturn in the telecommunications industry is forcing service providers to ensure that newly deployed platforms must be forecast-tolerant. That means next-generation optical switches must scale to very high capacities via an ordered and structured growth path. For instance, scalable optical switches must simultaneously support 2.5-Gbit/sec, 10-Gbit/sec, Gigabit Ethernet, and 10-Gigabit Ethernet traffic, yet also enable support for newer optical interfaces such as 10- and 40-Gbit parallel-very-short-reach (P-VSR) and G.709 digital-wrapper-based interfaces. The switch fabric must be highly scalable yet provide efficient modularity to enable a cost-effective growth path while maintaining redundancy and granularity down to STS-1 levels.
The goal is to combine these scalability requirements into a single NE to the maximum extent possible—simply parking smaller optical switches side by side negates much of these network values. Scalability does not simply imply terabit capacity but rather a smooth and structured evolution to this massive capacity so that capital/operating expenditures (capex/opex) are minimized, thus yielding the lowest cost per connect bit.
Network topology consolidation. SONET/SDH rings are predominant in the global public telecom network, since they offer extremely robust network topologies via several layers of protection. The scalable switch of tomorrow must allow for the consolidation of these existing rings into the optical switch particularly in the larger central offices that are strapped for floor space.
Collapsing rings into a scalable optical switch results in reduced floor space requirements, reduced power consumption, less NEs to manage, increased bandwidth management flexibility, and reduced equipment sparing. These achievable advantages are more crucial than ever to the short- and long-term economic viability of service providers in this hyper-competitive bandwidth market. Linear and mesh networks should also be supported by this massive optical switch so as to achieve improved economies of scale and more flexible bandwidth management.
Integrated ring and mesh functionality. Much discussion has taken place in the industry regarding mesh-network topologies and their apparent advantages over ring topologies. But to properly introduce mesh networks, a cost-effective migration path must be taken to avoid costly overlay networks.
It no longer makes economic sense to have ring-based networks and mesh-based networks operating in parallel, yet physically distinct from each other. A much more elegant and cost-effective solution is to consolidate existing bidirectional line-switched rings into the scalable optical switch, then build-out mesh networks as required using the same optical switch (see Figure 1). That mandates dual-functionality on the scalable optical switch.
The control plane for the mesh network must be based on standards instead of proprietary signaling schemes if truly global mesh networking is to be achieved. The reason is simple: Global mesh networks must enable mid-span meet at the control plane as well as the optical layer.
Signaling schemes based on one of the standards vying for global acceptance would allow carriers to interconnect to other carriers at the control plane as well, much like SONET enables mid-span meet at the physical layer. Proprietary signaling schemes will surely limit global mesh-network deployments by negating the possibility of dialing up bandwidth across different carriers using nonstandard proprietary signaling schemes.
Integrated DWDM. The value of integrated DWDM interfaces within the scalable optical switch is substantial from opex and capex viewpoints. The alternative to embedded DWDM interfaces is to physically separate them by housing them into collocated terminal bays. However, that greatly increases opex in terms of floor space, power consumption, and longer provisioning and troubleshooting times.
From a capex viewpoint, it implies additional redundant network equipment resulting in redundant electronics between the terminal and switch equipment, resulting in higher capex. A consolidated switch/terminal allows for efficient network management due to the unified nature of the NE itself—less overall NEs in the network to manage from an operations perspective.
A significant reduction in equipment sparing is also achieved. Figures 1 and 2 demonstrate the potential savings in combining the optical switch and terminal NEs into a single scalable optical switch. As clearly shown, the opportunity costs are drastically reduced.
Reliability and availability. Consolidating numerous disparate NEs into larger scalable switches means more traffic is handled by less equipment, thereby magnifying the potential traffic-failure scenario. But building on existing field-proven technologies, coupled with the redundant modularity of functional modules will alleviate these concerns in properly designed optical switches.
Improved ASIC technology also allows for consolidation of numerous separate devices in a single device to reduce silicon redundancy, lower power requirements, and increase the overall reliability of the optical switch itself. Reliability and availability have always been a major concern of service providers and vendors alike in the telecom industry and will be no different for these next-generation optical switches. Experiences gathered from the deployment of traditional SONET/SDH add/drop multiplexers and smaller optical switches will be crucial to the successful deployment of these large scalable optical switches of tomorrow.
The question is not whether photonic switches are required but rather when and where they will be required. Optical and photonic switches are simply not replacements for each other, as argued by some. These very different switches are highly complementary in nature and thus service different network needs in very different ways.
For instance, if sub-lambda grooming is required within a core switching site, then optical switches are the obvious solution. However, if sub-lambda grooming is not required at a given core-switching site for numerous co-propagating wavelengths, then the photonic switch is a better solution.
The choice of which switch to deploy is highly application- and traffic-pattern-dependent, although for the foreseeable future the inherent technical and economic advantages of the optical switch outweigh those of the purely photonic switch. A prolonged downturn in the industry, coupled with rapid advances in optical-switch technologies, will only serve to slow the wide scale deployment of these purely photonic switches—it's not a matter of if, but when.
Brian Lavallée is senior manager of systems engineering in the Nortel Networks optical Internet line of business (Montreal). He can be reached at firstname.lastname@example.org.
G. Simo, "Economic and Technical Strategies for Migrating from 10Gb/s to 40Gb/s Network Backbones," Nortel Networks, NFOEC 2001.