The key to boosting bandwidth in metro networks? * Business case * Latest developments * Typical applications
January 30, 2003
Interest in coarse wavelength-division multiplexing, or CWDM, is running high. On the one hand, there’s huge pent-up demand for a low-cost way of adding capacity to metro networks without having to lay fiber. On the other hand, CWDM has come of age in the past year or so – improving its potential for meeting this demand – not only in metro networks but also for other applications.
There’s little question about this interest in CWDM. In an ad hoc online poll of several hundred participants in the Light Reading Webinar that inspired this report, 67 percent of respondents said they expected four CWDM channels of 10 Gbit/s each to be the most cost-effective way of carrying 40 Gbit/s on a single pair of metro fibers within two years’ time; 37 percent said it would be 16 channels of 2.5 Gbit/s; hardly anybody thought it would be just a single 40-Gbit/s channel.
This shouldn’t be too surprising. Obvious, upcoming, big applications for CWDM are in metro access and in enterprise networks and storage networks. Already, customers are looking at 1-Gbit/s links here, and within a couple of years 10-Gbit/s interfaces will be available on switches and routing infrastructure. And CWDM allows carriers to respond pretty flexibly to diverse customer needs in a region of the metro where fiber is often at a premium.
What’s more, in principle, CWDM can match the basic switching capabilities of dense wavelength-division multiplexing (DWDM), but with the inherent tradeoff of lower capacity for lower cost. So there is no intrinsic reason why CWDM should not be a viable technology in terms of performance and price for the coming generation of reconfigurable optical metro networks based on DWDM wavelength switching.
So does this put CWDM and DWDM head-to-head in parts of the metro? Not necessarily, because the two have different roles to play that depend very much on carrier requirements and circumstances. But it does mean that carriers have to be able to disentangle the alternatives and decide which is the best for their needs. One of the aims of this report is to look at some of the common cases where CWDM really does have a winning cost advantage over DWDM or plain, brute-force, multiple fiber.
And even CWDM isn’t as simple as it was, as carriers can now choose zero-water-peak fiber to cram in 16 instead of the eight channels that standard singlemode fiber supports. So another range of options opens. And this isn’t just of interest to greenfield sites, as increasing amounts of zero-water-peak fiber are going into the meter. OFS, for instance, says that over 5 billion fiber meters of its Allwave zero-water-peak fiber have been installed to date, since shipping began in 1998, to more than 40 customers worldwide, covering metro access and cable TV networks.
For all these reasons it’s important to understand what’s new in CWDM technology, what it can do, and how the economics stack up for those crucial applications. And, of course, the most important thing of all – how you can use it in practice. This report gives a quick rundown on these topics with the aid of input from some of the leading vendors in the CWDM field.
Here’s a hyperlinked summary:
Reevaluating CWDM
Why CWDM has become a hot metro topic
CWDM vs DWDM
The pros and cons of the two technologies
CWDM Economics
The capex and opex savings that CWDM can bring
What’s New in CWDM
New standards and new fibers pack in more wavelengths
CWDM Building Blocks
Categorizing the components and systems architectures
CWDM Networks
Typical metro and data center applications
— Peter Heywood, Founding Editor, Light ReadingA lot of carriers are in a big dilemma. The amount of traffic on the Internet and elsewhere is continuing to grow as technologies like DSL and cable modems are deployed in access networks – but metro capacity is still limited in many places, creating a bottleneck that extends well beyond the last mile. So carriers need to boost the capacity of their metro networks if they are to handle this traffic.
But here’s the rub: Carriers are also desperate to cut costs to ride out today’s harsh financial environment. So they won’t add new capacity unless it is highly cost effective. And there’s new evidence that a lot of carriers aren’t convinced that the established DWDM technology is.Light Reading gathered this evidence in a recent wide-ranging survey of carriers and service providers in October 2002 (see Metro WDM: What Carriers Think), from which it is possible to assess the impact this conundrum has had on:
Their existing investments in metro DWDM/CWDM
Their attitude to future investment
The survey elicited 221 responses, 75 percent of them from carriers and service providers of different types from all over the world.
Right now, the survey found, Wavelength Division Multiplexing (WDM) – dense or coarse – isn’t widely deployed in metro networks. More than a quarter of respondents said they hadn’t deployed it at all, and another 27 percent said that they had installed it in less than 10 percent of their network.
Of course, this could be taken as good news: There’s a big market out there for a technology that wins wide acceptance by carriers.
Asked to rate on a simple 1-to-5 scale the issues that were discouraging them from deploying metro WDM, the high cost came out as the leading negative factor. Behind it, and scoring pretty well equally, were:
Lack of demand for high-bandwidth services
Availability of other technologies to solve problems
Immaturity of technology
Availability of dark fiber
All the same, plenty of respondents said they were planning to deploy metro WDM in the foreseeable future. 28% said they would be doing this in the next six months, and another 21 percent said they would be doing so within a year. Further ad hoc evidence for the growing interest in metro WDM came from the Light Reading Webinar on CWDM held in December 2002, when an audience poll showed that 43 percent of respondents expected CWDM to overtake DWDM in the metro.
So there is plenty of evidence that interest in metro CWDM is heating up. As Thomas Scheibe, product manager at Cisco Systems Inc. (Nasdaq: CSCO), points out: “More and more questions are coming from customers on CWDM and what you can do with it. The main drivers that I see are bandwidth needs in the metro as well as in the enterprise space. So customers who have fiber or who have access to leased fiber are really, really interested in deploying CWDM.”
Both CWDM and DWDM create multiple optical channels on fiber by transmitting different channels on different wavelengths, thereby allowing one fiber to do the work of several that support only a single optical channel each. But the two technologies are aimed at rather different requirements, and this creates some crucial differences.
DWDM
DWDM was really designed for long-haul transmission, where Erbium Doped-Fiber Amplifiers (EDFAs) are needed at intervals to boost the power of the light. EDFAs work over only a fairly narrow range of frequencies, known as the C- and L-bands (between 1530 and 1620 nanometers). So the wavelengths have to be packed tightly together.
In fact, vendors have found ways to pack channels really tightly together, cramming 32, 64, 128, or even more wavelengths into a fiber pair. And the fact that all the wavelengths can be given a boost of power from a single EDFA makes DWDM very cost effective in long-haul networks.
However, packing channels close together has a couple of downsides, especially in the metro world where the network economics are a bit different from those for long-haul networks.
First, you need to have high-precision filters that can peel off a specific wavelength without interfering with the neighboring ones. These are not cheap.
Second, you need lasers capable of keeping each channel exactly on target – and that nearly always means the laser needs to be kept at a constant temperature. In other words, cooling systems are required. High-precision, high-stability lasers are naturally expensive.
The bottom line is: high costs, high power consumption, and a big footprint, compared to CWDM systems.
CWDM
The whole point of CWDM is that it’s designed for shorter distances, where EDFAs aren’t required. As a result, it uses a much wider range of frequencies – 1270 to 1610 nm – and it spreads the wavelengths a long way apart.
Initially, channel spacing wasn’t standardized; but now it is, at 20 nm. This allows for wavelengths to drift about a bit, as lasers heat up and cool down. As no cooling is required, CWDM equipment is considerably smaller than DWDM gear. And it’s also a lot less expensive.
CWDM naturally doesn’t stretch very far, since many systems are unamplified, which keeps down costs, and very long spans are not a common metro requirement. Vendors cite distances of 50 to 80 km, although distances of 160 km or more have been achieved using amplifiers.
The main downside is that CWDM doesn’t support a lot of channels. Right now, eight is the usual maximum, although 16-channel systems are becoming available. However, whether carriers currently want larger channel counts is questionable, as the Light Reading survey indicated that most of today’s metro WDM systems seem to be running half empty (see Figure 2). The results were broadly the same for metro and enterprise deployments.
What seems to be happening is that metro carriers prefer to start with a small handful of wavelengths, while retaining the possibility of later expansion as demand or applications require. So carriers may plump for an initial two to four wavelengths on an access ring to handle the immediate demand for sub-10-Gbit/s access, and then scale up later.
Says Johan Sandell, VP of R&D at Transmode Systems AB: “We have a 16-channel system and it is upgradeable channel by channel from one to 16 channels without needing to break the traffic. Most of our customers deploy two to eight channels as a start, but they do, however, want to have access to upgrade capacity. And some of our customers are targeting this higher capacity from the start.”
A carrier’s specific circumstances play a key role here. Some consider that anything above eight wavelengths really belongs to their metro cores, for which DWDM, with its greater eventual capacity, would be the way to go (but see page 6 of this report for "DWDM over CWDM for Further Scaleability"). Others, with a shortage of fiber, may be attracted to higher-count CWDM systems from the start.
Santanu Das, director of metro optical system engineering at OFS, points out that one advantage of higher channels is that if there is a single-fiber ring rather than a fiber pair, both the working and the protection channels can share a single fiber by using double the number of wavelengths. “So particularly CLECs who need fiber can thus reduce fiber costs with higher-wavelength-count systems,” he says. “And facilities-based providers can generate additional revenues from leasing surplus fiber they don’t use with higher-count systems.”
At the end of the day, one of the big, simple arguments for CWDM is cost – it is potentially cheaper than equivalent multi-fiber or DWDM solutions. But how much cheaper? And under what circumstances? To throw some light on these issues, Light Reading asked OFS and Transmode Systems to present some results of their joint network modeling of CWDM systems. This looked at four network scenarios.
Four CWDM Scenarios
Figure 3 (below) compares four popular options for metro access networks, in which unprotected channels are assumed for simplicity. Two fiber types are used: ITU G.652 (conventional singlemode fiber – SMF) and the new G.652.C (low-water-peak fiber – see page 5). SMF typically has a high water-peak loss in the 1400nm region, restricting the number of CWDM channels that can be transmitted. In contrast, G.652.C enables transmission of 16 or more CWDM channels.
Case 1 shows each fiber type carrying a single wavelength per fiber, so one fiber pair is needed for each bidirectional fiber traffic channel. This is equivalent to adding a Sonet (Synchronous Optical NETwork) and SDH (Synchronous Digital Hierarchy) system on separate fibers.
Case 2 shows an eight-channel CWDM system with eight wavelengths per fiber; both fiber types work equally well here.
Case 3 extends the number of CWDM channels to 16, going from 1310 to 1610 nm at 20nm spacing; this can be done only over the G.652.C fiber.
Case 4 is the well-known DWDM alternative for 16 or more channels using the C- or L-bands, with wavelengths typically spaced at 200GHz for metro applications. Again, both fiber types are suitable here.
Figure 4 (below) is not to scale, but it serves to give a conceptual breakdown of the fixed and variable costs in the network.
The fixed portion of the network cost before lighting any of the fibers includes construction or cable installation costs like digging, trenching, etc.; it also includes cost of fiber cable, splicing, and connecting. For simplicity, cable installation is an ignored cost, as it can be quite variable and skew the results.
Optical line costs also have fixed and variable components, but the model amortized fixed costs over equipment modularity.
As channels are incrementally lit, the result is a staircase cost function, which is smoothed into a straight line in the subsequent graphs for simplicity’s sake.
In Figure 5, Case 1 (below, right) compares the cost of CWDM over single fiber against that of single-channel Sonet/SDH systems over multiple standard singlemode fibers. The x-axis shows the buildup of the number of channels. At the maximum number of channels (192 in this case) all fiber pairs will be exhausted.
Case 3 (left) assumes G.652.C fiber, and hence can support 16-channel CWDM. The y-axis shows the related cost normalized to the fully built-out CWDM case at 100 percent; no discount has been taken for the cost of capital.
There are several bottom-line conclusions that may be drawn here:
Because of the higher first cost of fiber cable, the initial investment prior to lighting channels would be approximately 15 to 20 times more expensive for the SMF solution compared to the use of CWDM.
At a low channel counts (say, 48), the single-channel solution is approximately four times more expensive than the CWDM solution.
When the cable is fully exhausted, the single-channel solution is still 75 percent more expensive than the CWDM case.
This suggests that CWDM is a very worthwhile proposition for metro build, since it can significantly drive down the cost of transmission without exhausting fiber plant.
City Reality
The theory looks attractive, but how does it work in practice? The CWDM analysis continues with a look at a real application in a medium-sized European city with a metro backbone ring running close to its downtown and financial district (see Figure 6 below).
A typical distribution of services here includes Gigabit Ethernet LANs, SANs, dedicated wavelengths, and SDH. Traffic requirements for the ring are 14 wavelengths. An economic model was developed to compare the cost of DWDM over SMF and that of developing CWDM over G.652.C fibers.
Figure 7 (below) gives the results, which are pretty favorable to CWDM, as this solution is about 55 percent lower in capex compared with DWDM.
There are several factors contributing to savings from CWDM:
Transponder cost savings for CWDM are of the order of 50 percent, owing to the use of uncooled lasers and pluggable optics such as GBICs and SFF (see CWDM Building Blocks).
Similar savings apply to the CWDM mux/demux equipment because of the avoidance of having to have thin-film-filter tolerances, and also from the lower labor costs of assembly, owing to the lower angular sensitivity of the equipment.
There are additional optics savings from lower power and space requirements.
A further point is that CWDM’s cost savings over DWDM could be sustainable. “As volumes pick up, both technologies will decline in cost over time, but the relative cost differences are not likely to diverge,” says Santanu Das of OFS.
But don’t forget that all these calculations are for greenfield networks – the situation can be very different for a carrier with an existing infrastructure. If such a carrier has enough spare fiber, it would normally use it instead of going straight WDM, since there will be no need to add mux/demux equipment. If, on the other hand, a carrier doesn’t have enough fiber, then the issue of whether to go for CWDM or DWDM can become pretty complicated.
And that will depend on a variety of issues, such as number of channels or wavelengths needed, distances (whether you have to amplify or not), and required future scaleability. Says Cisco’s Thomas Scheibe: “This will all define whether you use CWDM or DWDM. There will be cases where CWDM will be cheaper than DWDM, and there will be cases where you can’t use CWDM.”
In the last few years, two related developments have considerably enhanced CWDM’s status as a technology: an International Telecommunication Union (ITU) standard and a new type of fiber (also now an ITU standard), with a wider spectral window to support 16-channel systems.
Full-Spectrum CWDM
In June 2002, the ITU finalized a standard for what’s been dubbed Full-Spectrum CWDM, a.k.a. Recommendation G.694.2. This defines 18 discrete channels, 20nm apart. A couple of the channels at the edge aren’t of much use because they suffer from something called the Raleigh scattering effect. To make full use of the other 16 channels special low- or zero-water-peak fiber must be used.
Zero-Water-Peak Fiber
Conventional singlemode fiber suffers from a big attenuation problem in the 1400nm region, where the loss per kilometer increases roughly by a factor of three to four compared to that at 1310 or 1550 nm. This peak, known as the water peak because of its physical origin, effectively puts a big hole in the 1300-1600nm spectrum window used in telecommunications.
Figure 8 (below) shows this effect. In black is the loss curve of SMF, which has a high loss around the 1400nm water-peak region owing to the high initial loss from hydroxyl ions in manufacture and hydrogenizing loss that develops over time as the fiber ages.
The red curve below the black shows the loss of a zero-water-peak fiber like OFS’s Allwave, which completely and permanently eliminates this source of loss, creating some 100nm of extra usable bandwidth. Compared to SMF, four to five extra CWDM channels are reclaimed by zero-water-peak fiber, and this is the reason that CWDM vendors are able to move from 8- to 16-channel systems.
Note the blue curve for fiber dispersion on the right-hand y-axis, which is the same for both singlemode and zero-water-peak fibers in order to maintain backward compatibility with the installed base of equipment. This allows for support of all G.652 applications, such as Sonet/SDH, Ethernet, Broadband, and passive optical networks (PONs).
Just to confuse matters, some vendors also refer to zero-water-peak fiber. Zero- and low-water-peak fibers both conform to ITU Recommendation G.652.C, but there can be differences in performance. “The key parameter to watch out for is hydrogenizing loss over time, which is why it is extremely important to have fibers with low concentrations of silica defects,” says OFS’s Das. He also points out that zero-water-peak fiber has a slightly lower attenuation that SMF in the 1300 and 1550nm regions as well, which can be a further plus for carriers.
Greenfield Gains From Zero-Water-Peak Fiber
Unsurprisingly, greenfield deployments can show cost/benefits in using a 16-channel system on zero-water-peak fiber as compared to eight channels over SMF. This is shown in Figure 9 (below), where the capacity or the total number of wavelength channels is fixed, so the zero-water-peak fiber uses only half as many fiber pairs as SMF does. The 16-channel savings result from lower overall fiber cable costs.
This fixed cost is shown in green as a percentage of the 8-channel investment, which includes both electronics and fiber cable. The top of the red trace represents normalized SMF cable costs, with related savings shown by the red band. So when four channels are lit, the SMF cable with 8-channel CWDM represents about 90 percent of the investment cost. However, OFS’s Allwave cable with 16-channel CWDM is only 50 percent of the total cost, implying about a 40 percent savings. When 96 or more channels are lit, the savings approach 10 percent. So, given that there is always a savings to be had of between 10 and 40 percent of the investment costs, it is more economical to deploy zero-water-peak fiber and 16-channel CWDM.
Sixteen-channel CWDM and zero-water-peak fiber can also tackle fiber exhaust, and Figure 10 (below) explores the economic impact.
The cable count is fixed (unlike in Figure 9, where it was different for the different fiber types) and assumes 12 pairs, with the cost shown in yellow. Electronics can be added, either as CWDM (the trace shown in blue) or DWDM (shown in red). If the fiber were SMF, it would reach exhaust after 96 channels (12x8), so one needs a DWDM upgrade at that point. This is shown in green, with a similar slope to the red trace. If the cable were zero-water-peak, the cost trend would continue as in the blue trace; the 16-wavelength case at full buildout is the baseline 100 percent investment case. The savings in using zero-water-peak fiber is quite pronounced, approaching 33 percent compared to a midterm DWDM upgrade on SMF.
Of course, customers with older SMF may be forced towards DWDM or to lighting another fiber after eight channels of CWDM are used up.
Practical deployment of CWDM systems involves both the basic system building blocks, such as transceivers, and appropriate network architectures for typical customer applications. As always, there are some pretty fundamental choices that carriers need to discuss with their suppliers.
CWDM Building Blocks
The basic building blocks of CWDM are:
Transceivers
Multiplexer/demultiplexers
Optical add/drop multiplexers
Optical amplifiers
Transceivers are the modules that convert electrical signals into optical signals. They come in a variety of form factors, the main ones being Gigabit Interface Converter (GBIC) and Small Form Pluggable (SFP). They also support various data rates – typically, 1.25 and 2.5 Gbit/s.
Next there are mux/demuxes, which combine different wavelengths into a single beam of light, and vice versa. Then there are optical add/drop muxes, which add and drop wavelengths to and from a bundle.
Finally, there are optical amplifiers to boost transmission distances. The only technology available for CWDM that has a bandwidth wide enough is provided by semiconductor optical amplifiers (including linear optical amplifiers from companies such as Genoa Corp.). Those available on the market will cover four channels, but multiple amplifiers can be used to cover the entire band. Currently, the longest ranges, of around 160 to 200 km, are still experimental.
Transmode’s Sandell points out that customer requirements in the metro area vary quite a bit, so his company has to build a lot of different network topologies. “This puts some requirements on link spans and power budgets,” he says. “So we have developed a modularity in reach, giving 40, 80, or 120 km; and we use amplifiers at the longest distance. It gives us is the possibility of cost optimizing each network.”
There are two main types of CWDM system: transponder-based and GBIC-based (see Figure 11).
The transponder-based systems typically connect to a switch or router by using 850 or 1310nm optics. They incorporate a transponder to convert these optics into the wavelengths used in the CWDM system.One of the big pluses of this approach is that the equipment can connect to virtually any type of equipment: Sonet muxes, Gigabit Ethernet switches, Fiber Channel boxes – you name it, it can connect to it. So it can handle multiple services such as video, voice, Internet access, LAN-to-LAN, storage... over equipment from many different vendors. Adherents say that this gives a flexible, fully managed transmission system with a very clear demarcation point and the ability to perform signal monitoring.
Transponder-based systems can also funnel different types of traffic into a single wavelength – for instance, shunting one Gigabit Ethernet connection plus a couple of OC12 connections into a single channel. This is known as "muxponding."
The downside, however, is that transponders aren’t cheap.
GBIC-based solutions eliminate the transponder by plugging a GBIC straight into the switch or router. So they’re lower cost.The downside here is that this solution only works with equipment that has a GBIC port – which can limit the gear that can be connected. On top of that, muxponding isn’t possible, so it’s only one protocol per wavelength.
However, Cisco’s Scheibe says that Cisco adopted GBICs because GBIC ports are widely deployed today (perhaps around 10 million of them) and they are shipping at high volumes, giving a good base for using GBICs to leverage WDM. Further, in the market for metro enterprise solutions, the need the muxpond multiple services to one wavelength is not the primary driver, because customers typically need only three or four channels.
“CWDM GBICs are GBICs like any other GBICs, from a switch perspective. They are completely transparent to the switch or Layer-2/3 configuration, and there’s no separate management necessary for these GBICs. It’s a plug-and-play solution,” Scheibe says – and a way of extending the service offering beyond simple transport to a Layer-2- or Layer-3-based service. Monitoring of optical link performance is also doable with GBICs, as diagnostics are included.
Putting the Blocks Together to Build Networks
One of the surprising things about CWDM is the diversity of network applications to which the technology can be applied. Figures 12 and 13 show just a couple of examples.
Figure 12 (below) shows a real application Cisco designed for a customer, and it shows that you can do pretty complex scenarios with CWDM, because of the flexibility provided by multiple wavelengths. 
Three customers, A, B, and C share a single 2-fiber ring on which six out of the eight wavelengths are used. Customers B and C just need 2-Gbit/s point-to-point links between their sites over the Layer 1 optical ring. Customer A’s needs are more exacting, embracing a redundant data center (4-Gbit/s Ethernet link) and dual homing of the four enterprise sites with each data center (1-Gbit/s Ethernet links).
Figure 13 (below) shows a scenario from Transmode wherein the customer is running out of fiber and uses CWDM for fiber relief. The customer uses equipment from different vendors for the data storage and the LAN-to-LAN connection.
Says Sandell: “In this case, you see one application where muxponding is economical, and that is where you have a lot of ESCON channels. These channels are rather low speed – 200 Mbit/s – so by multiplexing them you make a more economical solution.”
DWDM Over CWDM for Further Scaleability
There’s an easy answer to concerns over the scaleability of CWDM. Just eliminate CWDM wavelengths in the C band and replace them with a whole bunch of DWDM bandwidths. This way, you end up with the best of both worlds – low initial costs and no sacrifice in scaleability.
This is not just a nice theoretical idea, as customers have already deployed such systems. Figure 14 (below) shows the basic idea. This approach does not really depend on whether you are using GBIC or transponders. Two wavelengths in the CWDM spectrum are replaced with an overlay DWDM, either by cascading the CWDM and DWDM filters (as shown in the upper left) or by putting them in series on the ring and using a free CWDM channel.
“The nice thing about this one is you can extend you channel count on the same fiber from eight or 16 by adding additional DWDM channels in the middle, with eight DWDM channels per CWDM channel, so you can go up to 24 or 30 channels per fiber,” says Scheibe.
The same idea works with 16-channel CWDM, but with a greater range of upgrade options. Says Transmode’s Sandell: “We are doing basically the same thing with a 16-channel system and remove one of the CWDM channels, adding, for example, eight DWDM channels, which gives you a very nice upgrade path. The thing you need to keep in mind is that CWDM components are cheaper than DWDM components, so you must compare [using overlaid DWDM] to adding another CWDM fiber, and see which is sensible costwise.”
Next Stop: 10 Gbit/s... and the PON
So what’s next? Pushing CWDM systems to handle 10 Gbit/s is likely to be the next big step in the technology, although significant progress will continue in lowering mux/demux losses and in extending spans through amplification.
Vendors are already announcing uncooled lasers for 10 Gbit/s in the 1550nm region, and migration to the 1300nm region will follow. Says OFS’s Das: “I think that is very encouraging, because we have a real potential to reduce the cost, even for 10 Gbit/s, although dispersion and attenuation will be the limiting factors there.”
But, in application terms, the next step may bring CWDM closer to you – literally. CWDM’s cost characteristics make it viable for smaller distribution access networks serving small businesses and larger residential applications, such as Fiber to the Curb – precisely the last-mile area where fiber bandwidth is at a premium.
“In PONs, CWDM gives a very nice way of upgrading,” says Das. Wavelength services could be offered over zero-water-peak fiber on top of the standard ITU PON TDM/TDMA protocol. “One could, say, use six extra wavelengths between 1370 and 1470 nm in order to connect three businesses on top of the 12 residences that are typically connected by a PON.”
You May Also Like
