Mesh or ring: Which gives the best optical?

Despite all the recent hype regarding mesh architectures and the emergence of the optical cross-connect, ring-based networks will remain popular among service providers

A lot of debate is ensuing about the merits of mesh networks vs. ring networks, triggered by the promise of optical cross-connects from several vendors. The view most often heard is that mesh networks of cross-connects are the only means to provide scalability to meet fast-growing demands for data capacity.

Nothing could be further from the truth. The protagonists may be new, but the debate is an old one. Many of the underlying truths of network design have been overlooked by a new generation of network designers.

This debate started when Sonet equipment emerged a decade ago and continues still. Rings and mesh have alternated for priority. Sonet add/drop multiplexers (ADMs) effectively introduced rings to the network.

The function of an ADM is to provide cost-effective access to a portion of the traffic on the fiber pair that passes through a node. Rings of ADMs were used for the collection and distribution of traffic - at first mostly DS-1s but later DS-3s. The rings initially were connected to asynchronous digital cross-connects that provided grooming and consolidation. These were linked in a mesh of point-to-point connections.

As Sonet versions of digital cross-connects appeared, they began to be linked instead by rings at OC-48, taking advantage of the quality monitoring and fast protection provided by these rings. Some operators found that they could not deploy OC-48 rings fast enough to cope with growing traffic, so they moved to a mesh interconnection between the digital cross-connect. This allowed for more parallel connections between any pair of digital cross-connect nodes. Then came OC-192 rings, whose rate of deployment is outstripping all forecasts.

Therefore, the historical message is that the focus in Sonet optical networks has moved back and forth between ring and mesh configurations as technology has evolved (Figure 1).

Real or virtual? In a sense, the debate has been as much apparent as real. A mesh frequently is managed as a set of logical but rearrangeable rings, particularly for fast protection, while in terms of connectivity, a ring often is used as a virtual mesh.

Another area where differences can be more apparent than real is in the amount of spare capacity needed for restoration. In a protected ring, the spare capacity is nominally 50%, but the use of shared protection effectively can reduce that figure. For a mesh, the amount of spare capacity needed often is quoted as being below 30%, but in a typical lightly meshed network, the required figure rises toward 50%.

Viewed at a high level, the movement between ring and mesh simply reflects the cost trade-off (Table 1).

As new technologies appear and different stages occur in network growth, one or other topology finds favor at some level of the network, often alternating between levels. This is a cycle that will continue (Figure 2).

So much for the hardware. What about the software? The industry uses much mesh-based control software within its network management, sometimes with the twist that premium customers can gain partial access to the network management system to set up and tear down some of their private lines. Because Sonet ADMs traditionally had only time slot assignment and lacked the time slot interchange that is inherent in digital cross-connects, they did not have the flexibility to participate in this process and, therefore, the circuit provisioning software usually excluded them.

A consequence is that any circuits that include Sonet rings still can be slow to provision, leading to the view that rings inherently are unsuitable for networks that have rapidly changing traffic demands.

But experience elsewhere shows that to be untrue. In optical networks based on synchronous digital hierarchy (SDH), the international variant of Sonet, ADMs have, from the earliest deployments, included time slot interchange and are widely used as mini cross-connects, even with ADMs at rates as low as 155 Mb/s (OC-3).

It is common practice in the SDH world for the centralized management provisioning of end-to-end circuits to include two or three ADM rings, as well as some digital cross-connects.

Ring and mesh in DWDM Will networks based on dense wave division multiplexing (DWDM) be influenced by the ring vs. mesh history? Certainly the debate has broken out again, triggered by DWDM moving up the value chain from being just point-to-point transport toward being part of a managed optical layer.

A forecast trend is for IP data payloads to be put more or less directly onto DWDM transport, with a minimal amount of Sonet in the way, so there is no obvious reason why the swings between ring and mesh in Sonet should influence DWDM networks.

Irrespective of technology, networks continue to be driven by the same factors: cost, features and manageability. If in DWDM, both mesh and ring can bring complementary benefits to fit each of these factors, as they have done for Sonet, then both mesh and ring will thrive, maybe in different network areas.

DWDM networks started as point-to-point links, and there are now tens of thousands of these in use. With so many links in place, high growth and churn rates are outstripping the ability of operators to manually deal with the required changes in connectivity. Staff shortages, crossed connections, erroneous data entries and unreliable connectors are just some of the problems. This is almost a rerun of why Sonet digital cross-connects arrived on the scene. The obvious parallel is to introduce optical cross-connects (OXCs).

One possible solution The term "optical cross-connect" is usually a misnomer because only the interfaces are optical. Today the cross-connect switch core is electrical, just as in Sonet digital cross-connects - hence the classification optical-electrical-optical, or OEO.

What distinguishes the new generation of electrical core cross-connects from their Sonet precursors is the ability to provide line rate switching, in which essentially the whole payload is passed through the cross-connect without any demultiplexing. This means that the entire payload that was carried by one wavelength of a point-to-point DWDM line system can be steered onto any one of several other DWDM line systems, even from different vendors.

A new generation of OXCs is expected to be deployed next year in which the switch cores are optical, formed, for example, from tiny moving mirrors that are based on micro electro-mechanical (MEMs) machines technology. The mirrors deflect the light beam from one fiber into another, moved by electromagnetic or electrostatic forces that are under computer control from the network operator.

The OXCs that are based on MEMs and similar technologies are sometimes known as "all-optical" or "photonic" cross-connects. Their advantage is that they can more readily cope with different data rates and especially ever higher data rates such as OC-192 and OC-768. Their classification is "OOO" or "triple-O," which stands for "optical-optical-optical."

Given all this activity geared toward an optical or DWDM mesh, where does this leave DWDM rings? Can the concept of a ring of Sonet ADMs be extended economically to a ring of DWDM optical ADMs (OADMs), in which it is wavelengths rather than timeslots that are added and dropped?

Some would think not, basing their view on the false assumption that each OADM necessarily involves the cost and complexity of demultiplexing to Sonet level at each node or at least converting every wavelength's photons on the fiber into an electrical signal to be switched, then converting back to photons. In fact, just as there can be OXCs that offer photonic through-paths, so there is technology for remotely re-configurable OADMs with the same attribute (Figure 3).

A major cost advantage of such reconfigurable OADMs is that they do not need transponders - which tend to be the dominant area of cost in DWDM - on the through path but only for the add and drop channels. Indeed, as with the whole managed photonics theme, switching an optics path in bulk at the photon level inherently is more cost-effective than doing so packet by packet with electrons.

Time for technology Two examples of reconfigurable OADM technology have been offered commercially.

The first uses switches based on the selective heating of waveguides in a polymer substrate. These operate on the Mach-Zehnder switching principle. The incoming DWDM signal is demultiplexed, one wavelength into each switch, using thin film interference filters. The required wavelengths are switched aside to transponders for "drop" local access, then "add" wavelengths are switched in as replacements, and the whole set is re-multiplexed by more thin film interference filters and sent to line as a DWDM signal again.

The second example is within a reconfigurable OADM product based on liquid crystal arrays for switching. Here the individual wavelengths are separated in space rather than in different waveguides. The light beams from the input diffraction grating, one per wavelength, shine on the individual liquid crystal pixels and can be passed or diverted under remote management control for the channel add/drop process. Each of 32 wavelengths can carry traffic at up to OC-192 speeds through up to 10 nodes without regeneration. The switchable through-path is literally glass and air, extending up to typically 600 kilometers of fiber.

If the ring is 100% protected by having identical traffic pass both ways around it, the capacity is 32 x 10 = 320 Gb/s. If no protection is required, the capacity becomes 64 x 10 = 640 Gb/s, all of which can be accessed at any node, provided that the appropriate add/drop transponders are fitted. The protection can be enabled individually for each wavelength to provide differentiated quality of service.

This type of product looks and feels like a Sonet ADM but with more than 100 times more capacity. It is, at its best, in regional networks, between metro DWDM and the ultra long-haul line systems now coming onto the market. The flexibility of a 0.3 to 0.6 Tb/s reconfigurable ring offers manageability, load balancing and resilience to one of the most congested areas of the network.

There are dozens of claimed candidate technologies for the role of photonic switching in OXCs and reconfigurable OADMs, but only a handful can give the necessary performance in terms of reliability, low crosstalk and low pass-through attenuation. OXCs and reconfigurable OADMs differ in their detailed requirements, so the industry should not necessarily expect to see a cross-fertilization of technologies such as liquid crystal arrays in OXCs or MEMs mirrors in reconfigurable OADM, but it could happen.

Fast signaling of new connections OXCs and reconfigurable OADMs are about to take advantage of another development: the concept of a single control plane for managing the connectivity of data (IP or ATM) and of DWDM. The problem of getting network management systems from different vendors to interwork for Sonet has proved tough, so the response of the optical transport community has been to propose some solutions (Table 2).

As a discipline, signaling has an excellent pedigree in terms of standards for inter-working between vendors, and it brings the benefits of potentially near real-time control - provided that the capacity of processors and communications channels is scaled appropriately.

The work toward this goal is taking place in several bodies:the Optical Internetworking Forum (OIF), Optical Domain Service Inter-connect (ODSI), the Internet Engineering Task Force and the International Telecommunication Union. The first result is an emerging user-to-network interface 1.0 from OIF. This will be used to control OXCs in mesh networks and OADMs in ring networks, plus combinations of the two. The key differences from Sonet, apart from the huge increase in bandwidth, will be:

- The use of fast signaling rather than network management to set up and tear down paths

- Paths with availability that has been confirmed by topology discovery rather than by the more error-prone database of a network management centre.

The only other requirement is the availability of OXCs and OADMs that have remote reconfigurability and the detailed feature set to support it. Just as Sonet did, the DWDM network is set to take advantage of both ring and mesh topologies.