Optical WAN Service Options

Because of limitations in optical processing, memory, and synchronization technology, ONs will not be able to provide packet/cell services which meet WAN switching requirements in the foreseeable future (Fig. 3).

In the near term, there is only one practical candidate technology for wide-area ONs, that is WDM. WDM is lower-cost than OTDM and has proven itself over long distances. In addition, wavelength routing of analog circuits (carrying digital data) has been demonstrated in numerous ON testbeds and is nearing practicality. In the long term, OTDM may supplant or complement analog circuits because of its ability to support very-high-rate users and the potential benefits of enhanced digital services.

 

Figure 9 – Layering in the next generation WAN architecture

Because WDM technology is more mature, the current universal focus is on WDM analog circuits carrying digital data. Although analog circuits may also be used to provide an analog transport service, such links are typically of low quality in the wide area. In addition, current focus is on full-bandwidth circuits because of the difficulty of time-sharing wavelengths in a long-haul mesh network. There appears to be little motivation for the latter anyway, given the large WAN traffic and the ease of performing multiplexing electronically.

Three illustrative visions of how WDM circuit services may be used are shown in Fig.5.

Figure 5 - Generic BBN alternatives

In the first architecture, the ON provides circuits between SONET/SDH equipment, which in turn provide lower-rate circuits between ATM/IIP switches. The ON drops local wavelengths and passes transit wavelengths to reduce the SONET/SDH switching requirements. The SONET/SDH equipment is used to add/drop and switch information within and between wavelengths.

In the second and third architectures, the ON interconnects ATM switches with full-wavelength circuits, and there is no SONET/SDH layer. Here, the ATM/IP switches may be required to process more information than in Architecture #1 (each port of the ATM/IP switch terminates a full wavelength).

The difference between Architectures #2 and #3 is the size of the ATM/IP switches and the amount of information carried on a wavelength. If optical circuits are a precious resource, the logical ATM/IP network (which is embedded on the ON) will be sparse. This implies large tandem ATM/IP switches since traffic will have to multi-hop through many switches. However, if optical circuits are plentiful, the ATM network can be more connected, thereby reducing the multi-hop traffic through intermediate ATM/IP switches.

Ongoing research is investigating the capacity of ON mesh networks. One interesting debate is the usefulness of wavelength changers. Initial estimates are a 10–50 percent capacity improvement, depending mainly on the network topology and routing algorithm. At this time wavelength changers are very expensive, and their inclusion may reduce the number of wavelengths that could economically be supported by drastically increasing the cost of a network node per wavelength. Complicating the matter, the most practical wavelength changers are only transparent to OOK modulation for rates up to ~10 Gb/s. This potentially limits the upgradability of the network. The use of wavelength changers remains a controversial and open question.

In order to facilitate end-to-end routing of information, a layered multiplexing and switching of information for different communication sessions are necessary (Fig. 6, 9).

Fig. 6 - Layered multiplexing.  

Fig. 7 - Layered switching architecture

Figure 8- Layered traffic and statistical gain

Layered multiplexing involves the concentration of traffic from users into communication pipes of increasing capacity from one layer to a lower layer. Each layer represents a different technology and protocol type. When traffic volume is small, multiplexing is usually performed electronically, while for highly multiplexed traffic the trend is to use optical multiplexing such as wavelength division multiplexing. This results from the well-known capability of optics:

·        high capacity for communication

·        but relatively slow processing.

One common multiplexing hierarchy is shown in Fig. 6. The layers are the virtual circuits, which are multiplexed into virtual paths, then into SONET/SDH pipes, then into wavelength channels, and then into fiber links. These pipes provide capacity ranging from megabits per second or less for a virtual channel, to hundreds of megabits per second for a SDH channel, to hundreds of gigabits per second and beyond for a fiber when many wavelengths are multiplexed.

Switching can be performed at each layer as an interexchange for information carried by different channels at the same layer, as shown in Fig 7. These channels are first terminated at a switch interface, and processing related to switching is performed. There are several functions achieved by such processing. First, the channel identifier is mapped into another identifier if the channels are labeled inband, for example, the translation of the virtual circuit identifier (VCI) and virtual path identifier (VPI) for ATM connections. For physical circuits, this translation is not necessary. In either the virtual or physical case, a circuit mapping table has to be maintained which is updated when connections are initiated or terminated. Second, information may have to be buffered for the purpose of synchronizing various connected circuits across the switch and for alleviating temporary congestion at either the input or output channels of the switch. Third, and perhaps the most difficult, is the control of the switching fabric for routing information carried by many channels

The fundamental purpose of switching is the reconfiguration of routes and their capacities as the traffic demand across a network changes over time. This fluctuation is often characterized on many time scales, as shown in Fig 8. For TCP/IP over ATM, we may have time scales such as that for a cell (ATM layer), a packet (IP layer), or a burst (TCP layer). Over longer time scales, we have the call layer, as well as the semi-permanent circuit configured to carry multiple calls. Over time scales of months, transmission facilities may have to be reconfigured to handle differences in traffic growth for geographical areas, or disasters such as earthquakes and fiber breakage.

As we move along the multiplexing hierarchy shown in Fig. 6, a statistical averaging effect is achieved when many connections are concentrated into a channel. In essence the burstiness of the connections is averaged out. At the semi-permanent circuit level, the randomness of the call arrival process is also largely averaged out.