The big question

Carriers that want to build out their networks must first address another fundamental issue: What type of traffic will their networks carry - and why?

More fiber optic network operators today are asking the question, "How should I build my network?" They wonder if they should go with a traditional Sonet design or one of the new optical approaches. Only a year or so ago, this question never would have been asked because only one answer was available: Sonet. Carriers could arrange Sonet for dense wave division multiplexing if their capacity requirements justified.

Today, however, there are more possible answers, and getting to the right one requires answering another question: "What is the nature and quantity of traffic on this new network?"

Although these start-up carriers usually can answer many questions, this one too often is not one of them. It's not that these carriers haven't done their homework, but they first must understand why the nature of the planned traffic is important. This will help them to correctly assess the true nature of an anticipated market. A quick look at the basics of Sonet, ATM and IP, followed by an examination of DWDM, can help. Carriers then should consider how to address some fundamental issues in making applications out of these technologies.

What is the traffic? The business plan should know

Traffic is the key to building out a network. However, as previously mentioned, it is likely that a start-up carrier will not know what type of traffic its network will carry. In developing its original business plan, the carrier should choose its target market based on more than geography, number of rooftops or number of new businesses in its service market. It should decide based on the targeted traffic. Such a determination will indicate if the start-up is targeting IP traffic, voice, video or a combination of these. While this decision doesn't mean that the carrier will be able to predict the future, at least it will enable the provider do its homework by developing the parameters for its projected market. This determination also allows the start-up to develop a network that is consistent with its business plan.

Of course, the business plan also can allow for a second choice and for flexibility. These kinds of allowances won't make the business plan any less useful nor will they prevent the carrier from making meaningful design choices. On the contrary, such statements in a business plan will define the need for flexibility in network design rather than leaving it up to the technical planner, who might end up guessing what the real network requirements are without such guidance.

Table 1 summarizes the characteristics of the various protocols and their relation to various traffic types.

Design by type

After identifying the differences among the various forms of traffic and associated protocols, a carrier must determine what each will require in terms of network design.

All voice (or all video). As shown in Figure 1, traffic in an all-voice network is handled on a time division multiplexing (TDM) basis. A full-time, separate circuit is established via the Sonet drop equipment for each requirement. This is the traditional fiber network and is representative of the majority of the existing networks.

Figure 1 shows a DWDM network that uses a number of wavelengths. The Sonet units, which could be an integral part of the DWDM equipment, provide a transfer-on-fault capability that is one of the most important aspects of the Sonet protocol. This function, called automatic protection switching, allows a Sonet route to switch quickly (in less than 50 milliseconds) to a standby route in the event of a failure - either equipment of fiber - or of a deterioration of the signal level or quality. This capability typically is used with fiber rings (of either two-fiber or four-fiber variety) to form a geographically diverse, full-time protection system. Quality of service (QOS) is not an issue because with TDM, each circuit is assigned a full-time, end-to-end path.

Mixed traffic. In terms of what carriers are installing, the mixed traffic network may be the most important because it is capable of handling any type of traffic. However, this flexibility comes at a cost. If a carrier plans to use a mixed network design when data is the main form of traffic, then it will have to build an overly expensive network.

This extra expense is generated from two points, the "cell tax" and Sonet protection. The cell tax refers to lost transmission capacity and is largely a reflection of the relative inefficiencies of the small ATM cells handling large IP frames (Table 1). The cell tax also is aggravated by the limited choice of sizes of ATM circuits, which can create a mismatch between provided circuit size and the actual bandwidth requirement. When combined, these effects can cause a cell tax from 13% up to 20% or 25%.

The other contribution to this excess cost is automatic protection switching. While automatic protection switching provides for extremely robust networks, it is a completely standby approach. This means that half of the network is not normally carrying traffic. The use of an extra pair of fibers may not be an out-of-pocket cost (in that there are likely to be spare fibers), but the extra equipment required at each node is a real cost. A Sonet node's backbone equipment cost will increase by a factor of "n" (n being the number of provided wavelengths) in a DWDM/Sonet network.

The first choice in a mixed environment is likely to be a Sonet/ATM network as shown in Figure 2. This depicts one node on a fiber ring network. All the traffic is carried on ATM riding on a Sonet base. On the drop side of the ATM units (the customer side), a conversion protocol will be used to convert IP data streams to cell-oriented ATM streams. Examples of such protocols include classic IP over ATM, multiprotocol over IP and LAN emulation. ATM provides QOS, and Sonet still provides the automatic protection switching.

Figure 3 shows a more recent mixed traffic development that includes DWDM capabilities. In this figure, the Sonet equipment is separated completely from the DWDM equipment. This separation allows the assignment of different uses (and different speeds) to the separate wavelengths. Thus, as in Figure 2, the traffic being handled is of a mixed nature, including voice, video and data. However, with this arrangement, the traffic is segmented by wavelength. This allows for flexibility in marketing the wavelengths. Again, Sonet continues to handle the automatic protection switching capabilities.

This arrangement emerged in the early part of 1999. Although it is an improvement for networks with a great deal of current or anticipated data traffic, this type of configuration still has drawbacks. Note that each of the provided services still requires the full Sonet Layer 1 protocol and its associated expense. This Sonet equipment represents a sizable, unnecessary cost - particularly if the carrier is building a data network that is an expansion of an existing Sonet network or a new network that will carry some mixed traffic but is mostly targeted for data growth.

One approach to improving this cost picture is to move away from Sonet on the wavelengths that provide data only. This approach is illustrated in Figure 4. In this approach, Sonet provides the automatic protection switching function for all the wavelengths, and the "Sonet Lite" protocol that the router adds provides the needed framing functions but nothing else.

The first wavelength provides automatic protection switching. Likewise, the ATM equipment provides QOS over its own wavelength. The IP wavelengths then provide best-effort transmission, unless equipped with multiprotocol label switching (MPLS), which will provide some of the QOS functionality. It also should be noted that the lack of QOS in any of the non-ATM designs can be improved by over-engineering the network, but at the cost of adding capacity. How much Sonet Lite really saves is unknown because the Sonet Lite implementation still looks a lot like regular Sonet equipment.

All data. The following methods are designed to be highly efficient in transmitting large quantities of data, whether via IP or other technologies. Because they are all relatively new developments, they are more or less proprietary. Newcomers to the equipment vendor business have championed many of these approaches, but lately, many of the traditional telephone equipment and data networking equipment vendors also are becoming heavily involved. The unprecedented growth of the Internet and its associated data traffic is driving these developments at a speed probably never witnessed in either the telecom or data equipment business.

Figure 5 shows a "traditional" approach to handling an all-data load with a router/Sonet approach. This method - IP over Sonet - replaces ATM with the addition of a Layer 2 protocol to provide the data-link functions. The point-to-point protocol is used in this instance. In the Figure 5 arrangement, Sonet still is the underlying technology, and it provides the automatic protection switching capabilities. The Sonet equipment operates with traditional rings and with the current network in the same manner. QOS is not provided (unless, as previously mentioned, aprotocol such as MPLS is added). While this design offers an excellent way to build an all-IP network, and some service providers are building out such networks, it still suffers the previously identified high cost of Sonet as the Layer 1 protocol.

Two developing methods of handling an all-data environment, "optical steering" and "wavelength routing," are illustrated in Figures 6 and 7. With these methods, the Sonet equipment is reduced to Sonet Lite for framing and is consolidated in the routing equipment. These methods have been covered in the press, largely because both have products that are under development. It should be noted, however, that due to the "bleeding edge" nature of these two approaches, no standards have been developed yet except those that are being reused from older methods.

Two novel approaches

In the optical steering approach, the automatic protection switching function is performed in the optical domain. A new piece of equipment is inserted in the wavelengths' paths - or included into the DWDM equipment - to allow wavelength provisioning and protection switching of wavelengths. This will be called an optical cross-connect switch for the purposes of this article, but various vendors use different names. The device detects the loss of a wavelength and switches to available spare wavelengths on a hop-by-hop basis to restore transmission paths. In the format used for the other network designs, the first part of Figure 6 depicts a network node using this approach.

It may not be immediately obvious, but this approach also implies a change in the topology of the network. As shown in the second part of Figure 6, this preferred topology is a mesh configuration. Advocates for this approach claim a 30% to 50% network savings over a ring topology. They also claim that the mesh arrangement is a more "natural" layout because it reflects the way that networks tend to grow, and it is more adaptable to existing larger networks.

Vendors report that, in their tests, the automatic protection switching function, which can be activated via the use of the optically steered wavelengths, is as fast or faster than Sonet-directed automatic protection switching. As in the previous ATM examples, QOS is not provided.

Alternatively, the wavelength routing design, which is a little older than the optical steering, relies on a highly capable terabit router to perform the automatic protection switching function via IP techniques. The first part of Figure 7 shows the equipment arrangement and protocol stack. If a transmission path is lost, the router attempts to restore it by using a different, available path on a hop-by-hop basis. These routers have multiple line speed interfaces (OC-48 for now) and can interface with multiple DWDM units. These DWDM units can be arranged in the traditional ring topology or in a mesh topology The second part of Figure 7 shows the arrangement as a mesh configuration.

The wavelength routers can interface to multiple gigabit routers on the drop side to aggregate IP traffic. Reports indicate that the time to reconfigure in the event of a loss of facility is much longer than with Sonet - perhaps even to the extent of minutes to reconfigure. As before, QOS is not provided.

Finally, some answers

So having examined all these issues, how do you build the network? First, design a network that will serve the type of traffic that supports your business plan. If the plan calls for mixed traffic, then don't build a data-only network just because it offers the latest in technology. If your business plan is to capture only data - and a few carriers have taken this approach - then the issue becomes one of evaluating the available and developing technologies. It may come down to a choice between technology that could become obsolete or technology that isn't yet proven.

The last answer is embodied in the concept of risk. A business plan may focus on a given type of traffic, but it also should focus on every possible risk: the risk that the traffic focus will not develop; that the traffic mix will be different; that a selected design approach will lock the provider into one vendor; that selected designs will become technologically obsolete; that a selected design approach actually may not be developed, or at least not developed as intended.

Considering these risk factors will allow a start-up carrier to develop a much stronger and better thought-out business case. It also will allow the entrepreneur to better instruct his network planner. Such instruction will assure the selection of a network design that truly supports the enterprise and is reflective of the business plan.