I apologize for the non-chart format for the charts, but Mike left that out of his instruction sheet and I can't figure out how to get them to work.
Project Network--JDSU (cont.)
I (cont.). Basic Facts: Optical Networks Overview
Existing networks have been cobbled together Rube Goldberg fashion. Traffic jams and roadblocks abound as networks are retrofitted to carry massive amounts of data. With demand for traffic doubling every 90 days, narrowband cannot take the load. RHK, optics research boutique, calculates the total amount of network traffic will increase by 30 times over the next four years with most of the growth in packet traffic, which eats up bandwidth (RHK, Terrestrial DWDM Optical Components, 2/00; 12/99).
RHK’s figures indicate data has already overtaken voice as a proportion of total traffic in most US networks. In 1999 interexchange carriers experienced growth in voice traffic of 15% but 60% in data traffic; regional bells 10% in voice but 50% in data; internet traffic grew by 300%, frame relay traffic by 65%, ATM traffic by 50%, and private line traffic by 23%. And the trend is accelerating.
The old networks being refitted were built for voice, which runs in steady small pieces, unlike data. The Sonet (US standard for high-speed data fiber-optic transmission) or SDH (European standard) systems carry traffic along doubled-up rings of fiber optics. Half the system remains idle, as backup. If one ring cuts out, traffic is shifted to the other in 50/1000 of a second. Moreover, Sonet is speed specific and processes traffic electronically. Getting on and off the Net’s optical backbone takes four steps: First, packets get addressed in IP; next they are bundled for delivery in another protocol ATM (Asynchronous Transfer Mode); then they are converted for transmission over Sonet rings; and finally they are packaged by WDM (Wide Division Multiplexing) gear as wavelengths for the backbone.
Networks have been “throwing bandwidth” at the traffic problem, as Bruce Brown points out in his “Fat Pipes” series on The Fool. A single fiber strand, only a little fatter in diameter than a human hair, can carry more than 1 million voice channels (Epoch, JDSU/SDLI, 7/27/00). But it’s no longer possible to solve the traffic problem by relying on laying more fiber or on increased efficiencies of high performance semiconductors, on faster routers and multiplexers, or greater fiber channel counts (MSDW, The Optical Networking Report, 2/14/00). ATM’s electronic switches are slow, inefficient, expensive
--definite bottlenecks. It’s a little like barreling along the autostrada at 180kph in a Ferrari, queuing at a toll booth for half an hour, and then exiting to a dirt road.
(Fat Pipes series: core boards.fool.com;
edge boards.fool.com
data storage/movement boards.fool.com.
The big three—NT, LU, and CSCO—are presently tied to electronics and optoelectronics of the Sonet past. They equipped the networks of the last decade and have long-standing links to its customer base. SDH/Sonet will increasingly be pushed towards the outer edges of optical networks to provide management and aggregation facilities. JDSU is not a Sonet vendor.
NG networks must 1) generate bandwidth; 2) provision wavelengths and steer them; and 3) aggregate voice and data for transport. According to MSDW’s David Jackson, DWDM will generate the bandwidth; core optical switches and cross connects will handle routing of wavelengths in the core (like Cisco routers route IP traffic in a LAN); and at the edge packet and cell optical transport boxes will aggregate voice and data for transport on access and metro optical rings. And these tasks must be accomplished through an end-to-end solution that is reliable, efficient, and economically viable for the carriers.
Market segments and growth:
Optical networks operate in several domains: long haul, metro, undersea, and cable television.
Optical Equipment Market Sizes & Forecasts
Market:1999/2003/CAGR
Sonet:
7.7B/17.2B/22%
DWDM and
Optical
Networking:
3.1B/15.2B/49%
Optical
Cross-Connects:
NM/1.8B/NM
Metro/
Interoffice:
NM/1.1B/NM
(MSDW, table 4)
Optical Component Segment Growth Rates
Market: Est. CAGR 1999-2003/Est. 1999 market size
Terrestrial DWDM: 55%/2B
Undersea DWDM: 90%/1B
CATV: 20%/300m
Sonet: 45%/1.5B
Total: 58%/4.8-5.0B
(MSDW, tables 5 and 7)
The economics of optics:
Data yields significantly lower revenue than voice even though pure data traffic is taking up more and more of the available bandwidth. In order to turn data traffic into profitable business for the carriers, costs of deploying optic networks need to come down, and we can expect price declines of about 15 to 20% over next few years. So far, performance gains are outpacing price declines. Demand pressures remain strong. While the cost of a kilometer’s worth of gear that carries a billion bits per second of data has plummeted, from $1,000 to $100, doubling price/performance every 10 months, internet traffic has been doubling every 100 days (RHK, CIBC, MSDW).
RHK points to the economic tradeoff involved in NG deployment (RHK, 12/99): For example, ultra-long haul--using Raman amplification, introduced by SDLI for remote pumping applications-- extends the range before a signal needs amplification from 600km to 6000km, cutting the number of line cards and DWDM terminals needed (about 70% of system cost) by a factor of 5 or 6 (MSDW and CIBC). A ULH DWDM system along a 2,500 km route carrying OC (optical channels) 192 would eliminate 400 line cards from the route. That’s a savings of $40m in line cards alone. Fewer electronic conversions also mean fewer points of failure and make upgrades easier. In a ULH optical line, the systems can be expanded by adding cards at the end point rather than all along the route. (http://www.lightreading.com/document.asp?doc_id=2014&page_number=3)
There are two subsets of optics economics: the notion of bandwidth glut and the debate on cap ex spending. Both would impact JDSU severely if the naysayers prove to be Cassandras and arrest network effects.
Bandwidth glut: Robert Metcalfe scoffs at the whole notion: “Saying there is a glut of bandwidth because there is a lot of fiber is a lot like saying there is a glut of microprocessors because there is a lot of sand.” Fiber capacity does not equal operating capacity. There is a major difference between fiber in the ground (which many analysts focus on) and bandwidth in the network. RHK estimates that much of the fiber in the ground is not usable—it’s dark, in which case, operators need to install equipment to light it up. Or it’s in the wrong place or old. Sycamore’s Despande estimates that 35 to 40% of existing bandwidth is unusable due to network mix-ups.
TMF Otter reminisces that rumors of bandwidth glut have been around since the dark ages in optics—about four or five years ago. When Qwest started to lay 25,000 miles of OC-48 2.4Gbs along railroad tracks, people laughed. Yet Qwest’s fiber was used quickly as it came on line. Global Crossing tripled capacity between New York and London over the last three years and had the same experience. fool.
Almost all future communications growth will be in packet traffic that requires excess bandwidth. The submarine portion, which is currently straining and has yet to connect India or China, is predicted to see strongest growth (Goldman Sachs, 6/8/00). While periodic fluctuations in supply are likely to continue, the trend points to further expansion. TMF Otter attributes this, in part, to the explosive growth in corporate network applications and their increased dependence on WANs (wide area networks) and other storage area networks.
Last-mile solutions are also in the works and involve laying more fiber. High-tech sewer rats—robots nicknamed Sam for sewer access modules—are being deployed by CityNet Telecommunications in Albuquerque to deliver, via sewer pipes, dark fiber cable to buildings
The cap-ex debate: Paul Sagawa of Sanford Bernstein kicked off the debate, maintaining that cap ex could not continue to grow faster than carrier revenues. Only four carriers, for example, were cash flow positive during the first half of 2000. Paul Johnson at Robertson Stephens immediately issued a strong rebuttal. Carriers, he believes, are caught in the telecom version of the prisoners’ dilemma. If one squeals, they all must squeal, and they can’t trust their fellow inmates to remain silent. Carriers cannot forego cap ex because a competitor is sure to forge ahead and capture market share. Customers, Johnson argues, want more services to increase productivity in their businesses, these services require a NG network, and the old voice-centric network is being commoditized. The carriers have to “pay to play.” If carriers could spend $1 trillion over the past 20 years to keep up with the growth in voice traffic, Johnson reasons, they will have no choice but to spend a similar amount to support the new services of the broadband revolution. (http://www.siliconinvestor.com/readmsg.aspx?msgid=14534918)
Blake Bath, Lehman telecom analyst, disagrees, calling the spending growth “unsustainable.” According to his report, the telecom-service industry next year will spend one dollar on capital equipment for every two dollars it generates in revenue. That marks a dramatic increase from 1996’s 1:5 ratio and this year’s 1:3.
Critically, the case against cap-ex so far has focused squarely on the cost side—ignoring the benefit side of decreasing marginal costs and increasing revenues. True, capital spending is increasing, as a percentage of overall revenues. But, as Johnson pointedly argues, the conclusion of a marked slowdown does not necessarily follow. The real question is the rate of return on the investment. Because of cheaper deployment costs and service capabilities, optical equipment generally pays for itself in a relatively short time. ULH gear costs anywhere from $30m to $45m, compared to $200m to $220m for Sonet transport gear. Some metro equipment can pay for itself in as little as three months (MSDW). JDSU CFO Tony Muller maintains that to stay in business component manufacturers must dramatically reduce costs for their customers’ customers (DeutscheBanc Alex. Brown conference,11/13-16/00).
Moreover, the gross figures used mask what is happening. According to the Dell’Oro Group, third quarter sales of optical equipment sank 7% from second-quarter levels, by $387 million, to $5.53billion. LU’s sales were down 14%, Alcatel’s 19%, and Nortel’s 6% (http://www.thestreet.com/tech/networking/1180249.html) Working from a smaller base, however, Cisco and Ciena both showed sales gains to the tune of 17% and 25%, respectively. The flattening spending trend, however, aggregates disparate markets. Lightreading director of research Scott Clavenna comments on the Dell’Oro report: “The DWDM and Sonet/SDH] markets are on very different trajectories. Companies that don’t have Sonet [Ciena] look to be doing much better than those with lots of Sonet [Nortel, Lucent, Fujitsu, Alcatel]” (http://www.lightreading.com/document.asp?doc_id=2610).
Juniper’s Scott Kriens maintains that “spending is going to be reallocated away from the old networks and legacy networks. It will move into the new infrastructure. We may see some decline in the absolute but the real dollar spending is going to move into the [new markets].” Obsolescence is also part of the equation. Whereas central office switches used to depreciate over 20 to 25 years, now that time is as little as five. (http://www.redherring.com/industries/2000/1208/ind-bigtelco120800.html).
Cap ex patterns, when broken out by segment, do show that carriers have reduced spending on the traditional circuit switched voice network to focus on highest growth areas, such as broadband access and IP networks/optical. Moreover, having spent heavily on laying new fiber (85% of which is still dark), carriers will concentrate on equipment to light up these networks. Infonetics’ report on “Service Provider Core and Edge Hardware” (11/20/00) points out that while spending may slow in some segments, spending in other segments will accelerate. Revenues for service provider core and edge hardware totaled $3.96billion in 3Q00 and are forecasted to total $14.7billion for 2000.
Trends:
Everyone who follows the optics sector agrees that NG and future deployment will involve four converging trends:
1. deployment in metro and submarine
2. conversion to a mesh architecture
3. integration of single-function fiber optic components into multi-function micro-optic integrated modules and chips
4. development of optical switching and intelligent networking
Metro and submarine: Increased demand for services and lower transport costs, coupled with high elasticity of demand for services, are driving fiber closer to the end user. The long-haul market was the early adopter of optical fiber because of fiber’s relative cost advantages, high capacity and low attenuation, but deployment is getting economical at the edge. Three years ago it made sense to deploy fiber only at the center of the network where it could be amortized over many users; now demand for services and reduction in costs have made fiber deployments economical at the edge (Epoch).
Submarine is somewhat different. Demand is high for new underwater systems connecting India and China or boosting capacity on the US/South America or US/Europe route. But caution rules the day. Initial costs are phenomenal and repairs prohibitive. And the submarine market is the most supply constrained. Only two independent vendors—JDSU and SDLI—have met the strict qualification processes to sell into this market.
Mesh architecture: Systems vendors are now developing mesh networks, which differ from the point-to-point connections of the current Sonet ring networks by creating multiple paths through a network. In other words, a signal can get from A to B, by way of C or D, not just directly. This new architecture levels the playing field for startup system vendors (Epoch). This point is important to JDSU, which has close relationships with NG systems companies and with ODSI (NG’s Optical Domain Interconnect protocol group).
Optics imported the idea of mesh architecture from the enterprise data industry where IP routers are linked in a mesh format and data can travel from source to destination by a number of possible routes. Routers select the most efficient route based on distance or “hops” and network congestion. Whereas the Sonet ring system requires 100% overbuild, mesh needs 30% to 50% backup. There are fewer elements to provision or maintain and greater scalability. A key benefit of mesh architecture is its ability to provision end-to-end services remotely.
Multi-function optical integrated circuits. So far DWDM has increased capacity and speed. Actual networking functions with the exception of link restoration have remained in the electrical domain. Kleiner Perkins’ Vinod Khosla sums up the situation: “A lot of work needs to be done in the optical component space. Bandwidth isn’t everything.” Software is needed to better utilize the bandwidth available and take advantage of equipment that can cram more data onto the fibers (http://www.redherring.com/vc/2000/1103/vc-khosla110300.html).
Today, optics are, for all intents and purposes, dumb and used for transport only. Optical signals traveling across networks must be translated back into electricity whenever they need to be given instructions. They are then translated back into light for further transmission.
The move from dumb optical networking (simply increasing bandwidth capacity and speed) to intelligent optical networking is just beginning to take shape. Combining subsystems into integrated optical circuits on chips (OICs) will be the key to success here. Attention has been given in the press to the number of hires JDSU has made from the semiconductor industry. It has generally been assumed these engineers were recruited to improve manufacturing. Given JDSU’s relentless pursuit of increased functionality in its products, moving from components, to modules, to subassemblies, it is not a stretch to believe this talent will be put to work on OICs, not just production processes.
Hybrids like MEMs (micro-electromechanical systems which in optical form can switch entirely in the optical domain using mirrors) tie various modules and components together on a single board, reducing size and decreasing cost. Hybrids should eventually turn into integrated optical circuits or multiple modules on a single chip. A DWDM multiplexer (multiplexers combine several channels to be carried by one line or fiber), complete with source laser modules, all on a single chip, would effectively create an integrated optical circuit (OIC). Although, as Jozef Straus admits, they are not there yet, whoever gets there first will have a bankable advantage. By moving up the value chain and offering OIC subassemblies, JDSU would help its systems suppliers reduce their own r&d costs and allow them to concentrate on their core strengths of solutions and selling. The pull in this direction is strong and JDSU’s talk of partnering with NG customers is more than lip service for the press. It will be a prime way for JDSU to capitalize on network effects.
Optical switching and intelligent networking: Software, directing photons around and managing the demand load, will be where the rubber meets the road. Managing optical systems increases in complexity and places new demands on software platforms as the networks become more dynamic, which they will as they transition to a mesh architecture. (http://www.lightreading.com/document.asp?doc_id=2014&page_number=4)
Now lightpaths across networks generally stop and are reset at every router. Hybrid optical/packet solutions bring packet processing right into the optical core (Sycamore’s SN10000 and Ciena’s CoreDirector approach). These hybrids emphasize flexibility and interoperability. ULH capabilities can be added only when needed and optimized around different transport distances. Sycamore’s SN8000, for example, uses line cards to frame incoming ATM and IP traffic with Sonet frames, but then transports the data on wavelengths and has the intelligence to add/drop wavelengths within the network, a big plus since expensive ADMs are eliminated. (http://www.lightreading.com/document.asp?doc_id=2014&page_numer=5)
Interactivity, interoperability, reliability, and cost saving—all require a level of intelligence and standards not yet achieved in the NG networks. In the future these networks will need high speed IP routers that can be optically connected; programmable add/drop optical multiplexers; and flexible layered bandwidth management. As operators extend networks to each other’s borders and look for ways to interconnect multichannel optical systems, routing becomes paramount.
At the end of November, Cisco announced that it will lead development of protocols for the multi-gigabit channel switching technology—what Cisco calls the optical control plane (OCP or multiprotocol lambda switching). But network operators have been reluctant to adopt any protocol. “The proof is in the pudding,” and Cisco has as yet no commercial product based on OCP. (It has announced , however, interoperability testing results between the Cisco12000 and Ciena’s CoreDirector). And OCP is only one of four schemes (and the more ambitious since it treats the optical and IP as one large layer) under development to let the IP layer provision lightpaths. Sycamore Networks spearheaded the ODSI effort, a simpler protocol which keeps the IP and optical layers separate with the interact occurring through a signaling scheme.
MSDW optical networking equipment and components analyst David Jackson maintains that wavelength provisioning at the IP router layer represents “a huge leap forward in network evolution….And early movers providing intelligent switches and end-to-end network management are likely to gain a dominant market share.” As DWDM systems progress from low- to high-channel count, traffic management functions become mission-critical and make coordination between JDSU and, say, a Sycamore or Juniper imperative as well as a way to generate network effects. |
