7/98 Netwatcher on DWDM [ASND references]
cimicorp.com
Strategies
In a marketplace that's used to electrical multiplexing in forms ranging from TDM to ATM, optical
networking seems like a real stretch. To make matters worse, the marketplace is full of hype (no surprise
there, right?) about the possible scope of the optical opportunity.
We've had optical networking for a decade or more. The new wave of optical interest, if there's any
substance to it at all, has to present not just optical fiber utilization but some way of substituting optical
processes for electrical processes within the infrastructure. If that happens, then the equipment balance can
shift, and vendor opportunity with it.
Dense Wavelength Division Multiplexing is a topic we've explored in a "definition" sense in a prior issue.
What we're going to examine here is how the market for DWDM could develop, and where that might impact
the rest of the network product and service market.
The Value of Traditional DWDM
What DWDM does, for those who missed our last piece, is provide a means of multiplexing multiple optical
carriers onto a single fiber by separating each in wavelength (lambda, as it is called). "Normal" WDM would
provide a dozen or so lambda carriers per fiber, while DWDM could provide hundreds.
Each lambda in a DWDM system can look to an electrical layer (SONET, for example) like a virtual fiber. This
means that the capacity of the fiber is multiplied considerably-hundreds of times if there are hundreds of
lambdas. It also means that there are now two ways of multiplexing traffic up to higher density-using the
SONET hierarchy or using DWDM. Each of these DWDM facts has its applications.
Higher capacity through DWDM can reduce the unit cost of transport in networks. The cost of a long-haul
connection is in part the cost of laying the glass along the path. The more traffic the glass can be made to
carry, the lower the unit cost of bandwidth to each user. This isn't saying that doubling the lambda count
halves network cost; equipment, craft costs, and profits all figure into the total cost of service. Estimates of
how much bandwidth cost makes up of total cost vary depending on application and the orientation of the
estimator, and range from about 15% to about 45%. All that range is probably valid in some applications.
Some may think that the cost-reducing benefits of DWDM would be enough to justify it, but that's not the
case. Carriers can't necessarily sell all the capacity that DWDM could create, and there's no incentive to
invest in new technology to lower margins and profits. There are certainly carriers and situations where
DWDM capacity augmentation makes sense-on an undersea cable, for example-but the "more-bits"
argument isn't compelling market-wide at this point.
What may be compelling is the multiplexing issue. Today, most multiplexing in transport networks is done
through SONET, a complex synchronized set of standards that provide for the creation of successively faster
optical trunks by the combination of lower-speed ones. An OC-3 is roughly three DS3s, and four OC-3s make
an OC-12-then so forth up to OC-192 or better.
The problem with the SONET approach is in how it handles high-speed data traffic. A single 155 Mbps
(OC-3-level) data trunk could be combined with three "telephony" OC-3s to create an OC-12. As the data
needs expand, however, the introduction of a single data trunk is enough to move a SONET network up to
the next OC-level-which doubles the capacity. All of the SONET add-drop multiplexers in each fiber nexus
have to pass this mega-pipe through, since none can handle its data payloads directly. This adds cost and
complexity to all the components of the network.
What DWDM does is let these fast data channels travel on their own lambdas, bypassing the SONET
hierarchy of add-drop multiplexers. The introduction of a fast data trunk along a fiber route now has no
impact on the SONET electronics already in place-it goes over another lambda with its own electronics to
support it.
The reason this is important is the growth of data traffic. If we measure network traffic at the point of
origination and convert everything to digital form, voice or circuit-mode traffic still makes up over 80% of the
traffic in the network. Most of this, however, is local calling that never touches a backbone. On long-haul
backbones, voice is only about 55% of traffic today, and that number is shrinking. The introduction of this
kind of data volume in standard SONET terms would result in all manner of new high-level SONET ADMs,
possibly displacing some of the current gear. With DWDM, the new data stuff goes onto its own lambdas
and the other traffic and equipment is untouched. The most efficient way to add high-speed data traffic to a
fiber network is DWDM.
DWDM In Non-Traditional Applications
Most of the talk about DWDM has been focused on what might be called the transport space, or the use of
DWDM on long-haul fiber. While the bit-savings benefit of DWDM is greatest where the cost of the fiber
path end to end is highest, the multiplexing value of DWDM is also at least interesting even in shorter-haul
applications-like the outside plant network.
Linking customers to a serving office is the mission of the outside plant. The makeup of this part of the local
exchange network varies with the density of business users and the age of the infrastructure, but it is
common to have a combination of fiber loops to business, fiber feeds to remote concentrators, and copper
loop.
Introduction of high-speed access connections to anyone in this structure can mean impacting a lot of
equipment in place, and that raises the costs. In addition, if the equipment in place is an incumbent LEC's
outside plant hardware and the serving carrier is a competitive player, any strategy that would impact the
hardware may be flatly rejected, no matter what the service revenue stream might look like.
DWDM would allow new local access service to be provisioned over a lambda that didn't impact the
incumbent equipment, customers, or services in any way. At the very least, this could drop the incremental
cost of the new access connection considerably because it wouldn't require displacing capital equipment
already installed. At most, it might enable a competitor to offer service where none could be provided
otherwise.
Transparent LAN service is a good example. TLAN requires 10 Mbps connection in most cases, which is far
above the level that would normally be supported on any outside plant multiplexing equipment. If a fiber
strand ran right by the proposed TLAN site, it couldn't be used in traditional networking because the new
service wouldn't be compatible with the electrical hardware associated with the service on that strand. With
a lambda, that old service and the new TLAN are "ships passing in the night" and would have no effect on
each other.
As great as this sounds, it's not quite that easy. Most of the older optical gear isn't compatible with DWDM
lambda division-the old stuff slops over all the lambdas and generally messes things up. Thus, it may be
necessary to replace older optical gear to make fiber DWDM-compatible in the first place. If that first cost
can't be justified, it may be hard to bootstrap DWDM in. This is particularly true in outside plant
applications, because the financial benefits of DWDM are more limited in these applications owing to lower
traffic density.
The Key to Multiplexed DWDM
If multiplexing benefits are the primary drivers to DWDM both in transport and access applications, then it
should be clear that transit switching at fiber nexus points is the key to multiplexing. Paths created over
multiple fiber spans using DWDM have to be cross-connected between the fibers at the meet points. If this
is done at the SONET/electrical level, the cost of the transit equipment is high enough to limit the value of
DWDM overall. This, we believe, is why most DWDM value calculations show the threshold traffic level for
economical operation to be OC-48.
There are two ways of providing lower-cost cross-connect at the Lambda level, but both depend on the
proposition that the entire DWDM payload is transiting the fiber nexus, meaning that nothing is being
inserted there or removed for downstream delivery.
The first strategy is to devise a simple optical-electronic-optical connecting device that extracts the lambda
from the fiber to an electronic format more readily handled, provides any necessary digital re-shaping, and
then converts the result back to lambda form. This mechanism would also allow "lambda-hopping", or the
conversion of lambda values incoming to outgoing to avoid collision of two payloads that want the same
wavelength.
The second approach is to perform the cross-connect in the optical domain. This concept, an aspect of what
is called "photonic switching", would eliminate the need to convert to the electrical domain completely. Some
research centers believe commercial photonic coupling of this type will be available within three to five years,
and it may even be possible to perform a lambda change with the conversion to help manage wavelengths
more efficiently.
A similar problem exists for add-drop functions in DWDM. Today, the practice is to terminate all of the
lambdas each time an insertion, drop, or cross-connect is required. This increases the ratio of equipment cost
to fiber cost dramatically in applications where many such points can be expected, which would be the case
for local access applications like transparent LAN.
Photonic processing of various types is clearly a key to the widespread deployment of DWDM, because this
type of handling would reduce the equipment cost in a DWDM network. Without it, the tendency will be to
focus DWDM on long-haul transport applications where the ratio of fiber cost to equipment cost is high
anyway.
DWDM and Higher Layers
Obviously, the introduction of DWDM as a formal multiplexing strategy would impact other technologies
with similar missions. SONET, with a point-to-point electro-optical interface and an overlay multi-path
network architecture, would certainly be impacted quickly. If DWDM were to be made cost-effective at OC-3
levels, for example, it would be doubtful whether SONET would be useful at all.
The relationship of DWDM to higher-level protocols like ATM is harder to foresee. It is very unlikely that
users will be provided with their own lambdas for network-building any time in the next two decades, in our
view. The ratio of user capacity consumed to transport capacity deployed determines the value of
multiplexing, at least in part. If users continue to operate at sub-DS3 rates for a decade or more (which we
think is likely), there will be a need to concentrate multiple user flows to reach the threshold of DWDM
economics. ATM could provide that concentration, creating an ATM electrical layer between telephone
switch and IP service functions and the DWDM transport network. This would clearly squeeze SONET out, a
point we've made in past articles.
It's pretty clear that this is at least part of Cisco's and Ascend's vision of optical networking. What is not
completely clear is exactly how it will all come about. The service provider industry is moving away from a
regulated monopoly status to a competitive market. In such a market, investment must be linked to improved
profits, and carriers have a long depreciation cycle on existing network technology. The best hope to
introduce something new on a large scale is to link it to a new revenue source. We've heard dozens of
candidate sources in the past, including multimedia, but none have proved valuable enough to bring about
major network change.
It is the revenue source, not the technology, that will catalyze DWDM.
Down the Line
In our next issue, we're going to review the recent activity in the "SS7" space. Cisco has acquired Summa
Four, a small switch vendor with a strong SS7 focus. Ascend now proposes to acquire Stratus, a major player
in SS7 nodes and processors. What's happening here? We will also be sending out our vendor questions on
advanced ATM carrier switch products early in September, with an eye on publishing the first analysis in
early December. Vendors note: We will be publishing reports in roughly the order we receive responses.
CIMI Corporation's market report "IP VPNs and MPLS: Twin Keys to 21st-Century Public IP Success" will be
available on August 19th. Interested parties should contact us for an executive summary and order form.
Please note that this is a controlled distribution report, and you must qualify to receive it. Do not send
money unless you have been advised that your order is approved.
|
