06/01/2001
Explosive data growth is changing the landscape of the network. From the early days of the digital cross-connect system (DCS) that statically switches TDM electrical signals, to frame relay/ATM/IP that dynamically switch packets, the network has never been so hungry for bandwidth. This demand is encouraging a new breed of technologies designed to make networks faster, more reliable, and more scalable than ever before. One of these technologies is the optical switch.
The optical switch, by definition, switches optical wavelengths. While the traditional DCS takes electrical signals in and sends electrical signals out, the optical switch takes optical signals in and sends optical signals out.
The optical switch can be classified into two categories. The first category changes the optical signal into electrical and performs switching function before changing it back to optical. This is known as an optical-electrical-optical (OEO) switch, or opaque switch. The second category is a pure optical switch where optical-electrical-optical conversion is not necessary. This category is also called photonic switch, because it switches photons, or lights, rather than electrons in the opaque switches.
Opaque or OEO optical switches provide a very high degree of flexibility. The fact that switching is completed inside of the electrical domain allows for switching for bandwidth ranges from STS-1 to STS-768 and beyond. The ability to manage across all these traffic types allows for cost-effective grooming and multiplexing for sub-lambda signals. Full visibility to all payload and overhead is a requirement to effectively police and administer service policies at the network edge.
OEO switches could, however, become high cost centers in the network. Termination of the optical line results in significant capital cost, as well as real estate footprint costs and power requirements. In addition, scalability for OEO switches can be a challenge when dealing with switching requirement of hundreds of terabits per second in the future.
Photonic switches--also called photonic cross connects (PXCs)--allow switching at the photonic layer to be done at a very low cost. Photonic switches have a fixed cost per port regardless of the amount of bandwidth through each port, because they switch light and do not discriminate between one lambda or 100 lambdas in the light. In addition, photonic switches do not care if each wavelength carries 10gbps or 40gbps--i.e., they are bit-rate independent.
Opaque switches, on the other hand, have an increasing cost per port as bandwidth increases. For this reason, at very high bandwidths, the cost of switching photonically is very attractive compared to opaque switches. This same property of photonic switches carries forward to space and power savings, as both remain constant regardless of switched bandwidth.
In addition to capital and facilities cost savings, photonic layer agility provides a unique opportunity for nodal scalability. Since the bandwidth per port is virtually unlimited by today's standards, a single switch can allow scalability into the 100s of terabits per second. By creating higher bandwidths through each port through grouping or "banding" of wavelengths, the photonic switch allows for extremely high nodal scalability.
The drawbacks of photonic switches are their inability to perform many of the functions achievable with OEO switches. Grooming of bandwidth below the "light path" can only be accomplished with opaque switches. Also, the inability to monitor electrical payload and overhead makes service policy management by the switch itself less attractive. Nevertheless, the cost, scalability and agility benefits of photonic switches can still be very attractive, especially in the applications where there is no need for sub-lambda bandwidth management, and the pass-through traffic is in dominance.
The comparison of opaque and photonic switches is relative between the two technologies. For example, the scalability of an opaque switch can also reach tens of terabits per second, which will be enough for many applications. Similarly, protocol independency also can be a common feature developed in the opaque switch, though it may not be as easy to implement as in the photonic switch.
There are a number of photonic switching technologies, including:
MicroEletroMechanical Systems (MEMS)--MEMS-based switches use static electric or magnetic force to control tiny (in micros) mechanical parts such as mirrors. MEMS technology has been around for a number of years for applications such as accelerometers for airbags, pressure sensors and CD heads. If the mirror can only be in two positions--"On" and "Off"--it is called 2-D MEMS. If the mirror can rotate around two axes to take many positions, it is known as 3-D MEMS.
Thermo-optics (e.g., bubble)--Thermo-optics technology, like bubble switching, uses local heat to create a bubble in the intersection to deflect a light beam, originated from ink-jet printer technology.
Liquid crystal--Liquid crystal switching is done through manipulating (e.g., filtering and rotating) the polarization states of the light.
Thermo-optics and liquid crystal are 2-D technologies. For any switches that use the 2-D approach, scalability can be a major issue. An example of an 8-port switch would use 64 switching points or mirrors, a N[superscript 2] problem that limits the switches to small size. For a 3-D MEMS switch with N-port, only 2N switches would make a full nonblocking crossbar switch.
Two planes of N mirrors are needed because the position and angle of the mirrors must be correct for low loss. Connecting multiple smaller switches into a multistage larger switch is not a viable solution because the loss incurred at each stage makes it difficult to compensate without amplification or signal regeneration that significantly can increase costs. Interconnect complexity is also an issue for multistage switches. Thus, the large photonic switches that exist today are based on 3-D MEMS technology. The size of a 3-D MEMS switch goes beyond 1,000 ports, making it possible to switch hundreds of terabits in a second.
Until recently, optical networks were mostly static. Services have taken a long time to provision, networks have been susceptible to manual errors, and service providers have usually over- provisioned their networks to avoid congestion and provide carrier-grade availability. When bandwidth grows fast and unpredictably, however, there is an urgent need to switch optical bandwidth so that a light path can be set up and torn down when and where needed. The International Telecommunication Union (ITU, www.itu.int) architectural standard for automatically switched transport networks (G.ASTN) provides dynamic optical connectivity though automatic routing and switching of transport bandwidth.
With optical switches and G.ASTN, it is now possible to bring intelligence, such as connectivity, to the optical layer. This intelligence transforms today's static optical transport network into a dynamic, service-oriented platform and converts the optical core into a more robust network--one that delivers unprecedented revenue-generating potential to service providers.
Sustaining a simpler and less resource-intensive network with automation, the dynamic connectivity also leads to significant reduction in capital, operational, and maintenance costs for both service providers and their customers. The customizable applications and services supported by dynamic connectivity include managed wavelength services, end-to-end optical Ethernet services, storage networking services, managed fiber services and bandwidth brokerage services.
All of the technologies discussed here would fall under the optical circuit switching domain because one sets up a light path and transmits information before tearing it down later.
The next step is moving to a photonic packet switching system, where the photonic switch would be able to switch optical packets. These packets can be sub-lambda information. To achieve such a level of optical processing, the photonic switch basically needs to do logic and calculation that has been possible only with transistors in photons. This new paradigm is photonic computing, which may not be commercialized for another five or more years due to a number of technological as well as business barriers. An optical transistor, for example, is inherently much larger in size compared to an electrical transistor, and thus makes it very difficult--if not impossible--to put millions of them into fingernail-sized silicon as with electrical transistors.
Before the photonic computing vision is realized, however, it is logical to develop a network architecture utilizing both technologies, because opaque and photonic switches each have unique and complementary advantages. This can be done by creating an architecture that could support an interworking of photonic and opaque switching. The opaque switch would be used for its optimal functions--demarcating the network edge, performing finer grooming and enabling services. The photonic switch would be used throughout the core of the network where pass-though traffic is dominant, managing very large amounts of bandwidth all-optically. This would result in capital facilities cost savings as well as much larger network scalability.
Billy Wu is senior manager of technology marketing for Nortel Networks Corp.'s (www.nortelnetworks.com) Optical Internet business. He can be reached at binw@nortelnetworks.com.
No comments:
Post a Comment