Finding the metropolitan network architecture that fits

Th 09spr04 1

Emerging technology has its risks and rewards. As metropolitan area networks start to support multiservices, which solution is the best?

The tremendous growth of data traffic in metropolitan networks has many carriers seeking a transport technology to replace voice-oriented Synchronous Optical Network/Synchronous Digital Hierarchy (SONET/SDH). While some carriers are "window shopping" for the future and others are desperate to fix trouble spots in their network, almost all carriers want a multiservice architecture that delivers a strategic cost or scale advantage. Regardless of their objectives, carriers face a confusing array of new technologies ranging from a simple variation on the current SONET/SDH theme to the radical prospect of the all-optical network.

A handful of these emerging architectures are based on relatively established technologies and therefore more likely to find their way into networks over the next few years. These include dense wavelength-division multiplexing (DWDM) rings, Asynchronous Transfer Mode (ATM) rings, Internet protocol (IP) rings, and integrated systems.

The DWDM ring is the most visible of the new metropolitan area network (MAN) technologies due to DWDM's success in long-haul applications. Both long haul and metropolitan DWDM systems leverage the fact that large chunks of bandwidth--OC-12 (622 Mbits/sec) to OC-192 (10 Gbits/sec)--can be multiplexed and regenerated more cheaply at the optical level than they can electrically. Another advantage is that any element on the DWDM ring processes only the wavelengths that it needs to access. Therefore, the cost of each network element is somewhat independent of the aggregate speed of the ring. Th 09spr04 1

Fig. 1. DWDM systems accessing many wavelengths (C) can coexist on the same ring as smaller systems accessing one or two wavelengths (A and B).

Larger DWDM systems that access many wavelengths can coexist on the same ring with smaller systems that only access one or two wavelengths (see Fig. 1). This capability also results in lower network cost for cases where the aggregate speed of the ring is very high, because each element only incurs the cost of processing the bandwidth accessed. SONET/SDH, on the other hand, requires that every network element electrically terminate the entire bandwidth of the ring to perform the time-division multiplexing (TDM) function.

Most major equipment providers, as well as a host of startups, have proposed DWDM ring systems. These systems vary widely in their function. Some simply provide raw wavelengths, or "virtual fiber," and require external equipment for formatting and management to make the bandwidth usable. Other systems provide integrated performance monitoring and may even format the bandwidth into OC-12 or OC-3 (155-Mbit/sec) line rates. Some systems allow flexibility for deployment of DWDM mesh or star topologies on top of the physical ring infrastructure.

Regardless of the DWDM ring implementation, a major drawback is that each node pays a "penalty", even if only one wavelength is added or dropped at the site. That's because, in order to participate in the DWDM ring, the optical add/drop multiplexer (OADM) must "pick off" a wavelength tightly packed within the 1550-nm range with other wavelengths on the ring. This pick-off requires a sophisticated optical multiplexer/demultiplexer function and long-reach 1550-nm transmitters and receivers with wavelength stabilization capability. In a system not using DWDM, the alternative is less-expensive, generic 1310- or 1550-nm optics and broadband receivers. As a result, sites with low-bandwidth requirements on a DWDM ring pay a high cost to participate with higher-bandwidth sites.

Even with this penalty, DWDM rings are more cost-effective than SONET/SDH rings for delivering large amounts of bandwidth, especially if there is a requirement to drop an OC-12, OC-48 (2.5 Gbits/sec), or Gigabit Ethernet at every site on the ring. But DWDM rings are not always cost-effective in situations where some sites need high bandwidth and others do not. For example, if a five-node ring has two sites where an OC-12 is dropped, one site that needs an OC-3, and two others that require a few DS-3s (44.736 Mbits/sec), all five sites must still pay the DWDM penalty to participate in the ring. Economically, this drawback makes it difficult to justify DWDM rings in networks where bandwidth requirements are uneven or uncertain.

ATM rings have evolved as an answer to address the inefficiencies of carrying nonchannelized data over a channelized SONET/SDH infrastructure. The rates assigned to virtual connections in an ATM network have arbitrary granularity and can thus provide much better utilization. Taking full advantage of the statistical multiplexing capabilities of ATM can improve the utilization even more, especially in the presence of large volumes of bursty data traffic. Synchronous traffic such as voice T1 (1.554 Mbits/sec) and T3 (44.736 Mbits/sec), can also be carried over ATM using its circuit emulation mechanism.

These properties have resulted in the deployment of an ATM network that provides grooming and service adaptation, overlaid on a SONET/SDH network that offers reliable transport. ATM rings are an attempt to provide the protocol's benefits directly over a fiber ring, without the SONET/SDH transport layer. Th 09spr04 2

Fig. 2. Each node on an ATM ring (shown here) functions as an ATM switch or crossconnect with two ports on the ring and one or more ports for local traffic. ATM traffic coming in through one ring port is diverted (dropped) to the local port or sent onwards in the ring through the other ring port. Traffic decisions are made independently for each incoming cell, based on its virtual-path/virtual-circuit indicator.

Each ring node functions as an ATM switch or crossconnect with two ports on the ring and one or more ports for local traffic (see Fig. 2). ATM traffic coming in through one ring port can be either diverted (dropped) to the local port or sent onward in the ring through the other ring port. The decision is made independently for each incoming cell, based on its virtual path/virtual circuit indicator (VPI/VCI). Similarly, ATM traffic coming in through the local port can be sent to one or both of the ring ports, and in that case, it is merged (added) with pass-through traffic from the other ring port. Any local non-ATM traffic is converted to ATM. Virtual connections are provisioned by programming the VPI/VCI routing tables in the ring nodes; each connection can have arbitrary bandwidth.

Because of their ability to groom data traffic to virtually any granularity, ATM rings provide much better utilization of bandwidth than SONET/SDH when data is the primary application. In addition, ATM provides standardized adaptation of data interfaces such as Ethernet, to allow a wider variety of data services.

The disadvantage of ATM rings is that they manage bandwidth only at the electrical level and therefore do not provide the scale and upgrade capacity of DWDM rings. Most of the proposed ATM rings run at either OC-3 or OC-12, which means they are poorly suited to terminate large amounts of DS-3 or OC-3 traffic and cannot handle termination of OC-12-level traffic at all without future upgrades to OC-48 ring speeds.

IP rings are similar in theory to ATM rings but use IP routing technology instead of ATM for multiplexing. These rings are generally based on generic routers with interface cards that allow them to connect directly to an optical ring, bypassing the need to run over a SONET/SDH network. For networks that have access to fiber and consist only of routers, IP rings offer a viable alternative; the router is considered a "sunk" cost and the interface cards are much cheaper than a complete SONET/SDH system. Interface cards, however, combined with the cost of a gigabit router can be very expensive compared to pure ATM or SONET/SDH transport technologies. The best case for IP rings is in situations where very large routers are already required.

Proposed IP rings run at OC-3 to OC-48 line rates; so like ATM rings and SONET/SDH, that architecture cannot offer the same scalability as DWDM rings. In addition, unlike ATM rings, IP rings are not well suited to carry traditional voice, leased-line, and other circuit-based traffic. The best scenario for IP rings is configurations where large routers have to be installed anyway, services other than IP are not needed, and growth that requires DWDM scalability is not anticipated.

A few systems are now emerging that combine multiple technologies such at TDM, ATM, IP, and DWDM into a single platform. The advantage of integrated systems is that they can offer a wider variety of service interfaces (ATM and SONET/SDH) and the scale of DWDM rings, and they can groom traffic to more efficiently utilize wavelengths. As a result, these products are targeted at networks that need to de liver traditional low-speed circuit services and high-bandwidth data services over a common network infrastructure.

Some of these products consist of TDM platforms with DWDM or ATM plug-in cards. The systems designed from the ground up as integrated platforms, however, take greater advantage of integration. These systems can offer multiple services, traffic grooming for data and voice, the scalability of DWDM, and reduce the number of network elements. Integrated systems also lower costs, removing the expense of individual network elements. Seamless management and provisioning of the data, transport, and optical layers of the network is another benefit, reducing operational cost and lowering service provisioning time. Th 09spr04 3

Fig. 3. The functions of multiple network elements and layers are provided in an integrated transport element.

An integrated transport element can provide the functions of several network elements and layers (see Fig. 3). Depending on the implementation, any service--from circuit to ATM to wavelength--can be offered over an optical transport system running at any bandwidth serviced by a single network element.

While an integrated system can help lower the overall system cost by making better utilization of wavelengths, the use of an integrated DWDM ring still incurs the DWDM penalty. Every network element must pay the high cost of DWDM optics even if a particular site does not need it. To eliminate this problem, at least one integrated system offers the ability for network elements using DWDM to be combined on a ring with network elements not using DWDM, working together as a single system. This allows both lower-speed services and higher-speed services to be supported on a single ring, without the DWDM penalty at the sites that do not need DWDM bandwidth.

Pure DWDM rings are attractive if the market is wholesale wavelengths or pure OC-12 and OC-48 circuit transport. In these applications, the high first-cost of DWDM rings is justified. ATM rings address the needs of low-speed access networks when the services offered do not surpass the T1 to DS-3 level (with occasional OC-3), and aggregate bandwidth is not expected to increase beyond OC-12, or OC-48 if available.

IP rings offer a viable solution when large routers are already necessary, only packet-based services are anticipated, and the growth capacity of optical networks is not required. For situations where the service provider wants to offer high-bandwidth data and traditional circuit services as well as have the flexibility of DWDM, an integrated system is the best solution.

Doug Green is vice president of marketing at Chromatis Networks (Bethesda, MD).

More in Network Design