Implement 10 Gbits/sec now, 40 Gbits/sec when it makes sense

Th 79344

SPECIAL REPORTS: Annual Technology Forecast

In 2002, 40-Gbit/sec technology will cost 10 times what a 10-Gbit/sec installation costs for just four times the throughput.

TOM MOCK, Ciena Corp.

The question of whether to deploy a 10-Gbit/sec system, operating at 25 GHz, or a 40-Gbit/sec system at 100 GHz over the next year can be resolved in simple economic terms. The short answer is to opt for cost-effective, proven 10-Gbit/sec technology in 2002 and plan to upgrade to 40 Gbits/sec in two or three years. Of course, most vendors' bottom lines would look better next year if their service-provider customers were pushed to buy lots of 40-Gbit/sec systems in 2002. However, over the long haul, pushing 40-Gbit/sec products into their networks in 2002 would not be healthy for either vendor or buyer.

Regardless of a service provider's long-term vision, there is no question that today 10 Gbits/sec buys more reasonably priced throughput than does 40 Gbits/sec. It certainly is more network-efficient in terms of power and footprint. Most importantly, 10-Gbit/sec technology is available for deployment and operational support now. Although several companies are working on 40-Gbit/sec systems for the future, in today's market, 10 Gbits/sec makes sense.

No matter where a service provider is in its deployment curve, the right approach to network growth is evolutionary, not revolutionary. The access rate does not need to match the transport rate on a backbone. Each should be groomed separately to optimize cost and convenience. Today, 10 Gbits/sec provides that rate economically. Given today's economy, the name of the game is to find something that moves data at the lowest possible cost. For the next year or two, that "something" is narrow-spaced 10 Gbits/sec.

There always will be a market for multiple channels of 10 Gbits/sec, and the network of tomorrow won't eliminate its 10-Gbit/sec trunks to install a fat 40-Gbit/sec pipe. That is especially true in cases where the lines extend over very long distances or in operations where the fiber plant is old.

Currently, 10 Gbits/sec is the baseline for tightly spaced, high-speed transmission. It is simple to demonstrate that 10 Gbits/sec with tight channel spacing at 25 GHz (or eventually, 12.5 GHz) will be more cost-effective and network-effective in 2002 and most likely through 2003. Th 79344

Expect that lowest-cost data rate, as shown by each point in the chart, depends on distance and capacity.

Long term, the question service providers face is not so much buying exclusively into 10 Gbits/sec or 40 Gbits/sec. Rather, it is deciding which technology makes sense for which segment of their networks. Just as carriers were hit with sticker shock when they first migrated from 2.5 Gbits/sec to 10 Gbits/sec, so they will face higher costs when migrating from 10 Gbits/sec to 40 Gbits/sec, both in dollar costs and hidden technology costs. There is no current compelling need for any carrier to move to 40 Gbits/sec. "Bragging rights" about being on the cutting edge of technology can prove very expensive. There is no reason to swallow those costs while the price tag on 40 Gbits/sec remains well above its eventual mass-deployment cost.

In 2002, 40-Gbit/sec technology, where available, will cost roughly 10 times as much as the same 10-Gbit/sec installation. Why pay 10 times as much for 40-Gbit/sec technology when it gives only four times the benefit? And why pay 10 times more for a product that is not yet a proven commodity?

Rather than jumping into 40 Gbits/sec, it is best to review the current market position of each technology and decide accordingly. There is no compelling need to buy 40 Gbits/sec simply for innovation's sake. People are pursuing 40 Gbits/sec today since they believe that down the road it will be less expensive on a dollars/bit/mile basis. It is important, however, that today's 10-Gbit/sec platform be positioned to allow migration to 40 Gbits/sec when it becomes economically viable. The rules of physics aren't just cool ideas-they're law. Engineers immediately recognize that most transmission impairments are a function of the square of speed. While 40 Gbits/sec does give four times the throughput, many of its transmission impairments are 16 times worse. Additional equipment required by a 40-Gbit/sec system to compensate for impairments may well dominate the cost issue beyond any transport savings, at least in the short term.

Reach also favors the 10-Gbit/sec system. In fact, reach provides another real advantage to 10-Gbit/sec systems. When transmitting a signal at a lower data rate, all else being equal, the 10-Gbit/sec signal will propagate farther. Granted, every network equipment provider has its own special compensating mechanisms, but any such fix that works on a 40-Gbit/sec system will also work on a 10-Gbit/sec system and can be produced at a fraction of the cost for the latter. Regenerator spacing will always be closer together for 40 Gbits/sec than for 10 Gbits/sec simply because of the laws of physics-there is a fundamental capacity-versus-distance curve. Th 79345

While it may be possible to make 40 Gbits/sec reach 2,000 km, the question is whether the dollars spent enable the realization of sufficiently high capacity over these long distances. To carry large amounts of capacity over distances beyond 1,000 km, there is a true disadvantage on a 40-Gbit/sec network.

Given that 40 Gbits/sec will not in crease the amount of capacity down a fiber compared to multiple wavelengths of 10 Gbits/sec, there will be little difference between deploying 320 channels of 10 Gbits/sec and deploying 80 channels of 40 Gbits/sec. Both require the same fiber capacity. While it might seem more efficient to equip just 80 channels, the much higher cost of the 40-Gbit/sec electronics quickly destroys any financial advantage.

Because they take advantage of mature technology, the narrow-channel-spacing products available in the near term offer as much capacity as "bleeding edge" 40-Gbit/sec products promise. Yet, 10 Gbits/sec does it with better economics.

There are costs incurred for initial procurement and installation of equipment and the ongoing operation of that equipment, so for 40 Gbits/sec to meet 10 Gbits/ sec on a level playing field, the cost of 40-Gbit/sec terminal equipment can not be more than four times the cost of 10-Gbit/sec equipment. Today, 40-Gbit/sec equipment costs approximately 10 times more money, especially when additional components required to address transmission impairments are considered. Unless the goal is to show off a "hero network," there is little reason to pay 2.5 times the money for the same basic capacity with an unproven product. Waiting until 40-Gbit/sec systems are more mature and lower in price may make more sense.

The economic parity on operating costs is similarly disproportionate. Those deploying demonstration networks with 40 Gbits/sec will gain bragging rights and establish an innovative leadership position. In today's market, however, they will have to justify the 40-Gbit/sec expense to their stockholders as test-beds rather than for their operational economic benefits.

Another economic problem comes from the installed base of fiber. Certain fiber types, including commonly deployed fiber, simply do not work well with 40-Gbit/sec systems. The characteristics of any fiber with significant dispersion work against efficient long-haul use of 40 Gbits/sec. Because it is 16 times more sensitive, it is much more difficult to compensate for that impairment.

The polarization-mode dispersion (PMD) tolerance of a 10-Gbit/sec system at 12.5 GHz is 16 psec. With a 40-Gbit/sec system at 50 GHz, PMD tolerance is 4 psec. In fact, 40 Gbits/sec will require PMD compensation at many distances where it does not come into play in 10-Gbit/sec networks.

PMD sensitivity increases as the square of the data rate. This propagation effect spreads out the waveform and creates a need for dynamic compensation. Thus, a PMD compensator will be required for most 40-Gbit/sec systems. Consensus exists among engineers that standard chromatic dispersion compensation techniques used in the market today will not be effective enough for 40-Gbit/sec systems. They will require dynamic chromatic dispersion techniques as well. That eventually will become a commonly available device, but it will not be inexpensive.

Today, engineers deploy dynamic chromatic dispersion devices across the entire transmission band of a 10-Gbit/sec network, or at least across a sizable number of channels. With emerging 40-Gbit/sec systems, the dynamic chromatic dispersion devices will need to be deployed on a channel-by-channel basis. The differential cost in control electronics for the add-on boxes must be added to the cost of fixed components, and the bottom line will be substantial.

Channel cards for 40-Gbit/sec networks already are more expensive, thanks to their higher-speed electronics. With the addition of dispersion devices to that cost, no wonder there is real fear in many boardrooms that the compensation issue will have a significant-if not dominant-effect on the cost of tomorrow's 40-Gbit/sec networks.

Today, 10 Gbits/sec offers a +32-km non-dispersion-shifted fiber (NDSF) dispersion tolerance. With 40 Gbits/sec, expect to see that figure plummet to just +2 km. Again, the squaring effect, a factor of 16, rears its ugly head. It must be possible to compensate a long piece of fiber to within a 2-km distance. On a 1,000-km link, compensation must be just 0.2%-a difficult task. Not only must the actual dispersion value be accurate for the fiber, but installers must also get their dispersion slope values correct. Even a slight variance in slope over a long distance begets a substantial difference in dispersion.

In the 10-Gbit/sec world, tolerances in dispersion due to temperature or other causes are negligible. However, as faster systems move the dispersion tolerances down the curve, that becomes less true.

One disadvantage of narrower channel-spaced 10-Gbit/sec systems is the necessity of tighter channel frequency control. Since a comparable 40-Gbit/sec system would have channel spacing four times as wide as that of the 10-Gbit/sec system, the frequency accuracy of 10-Gbit/sec systems will need to be four times better than that required of its 40-Gbit/sec counterpart. Fortunately, that is not really a major hurdle with 10 Gbits/sec, since the technology to correct the problem is currently available.

It is reasonable to expect the first 40-Gbit/sec systems to use 100-GHz spacing. That is comparable in density to a 25-GHz 10-Gbit/sec product. The practical limit for channel spacing in a 40-Gbit/sec system is 50 GHz. It might appear that a 40-Gbit/sec system is more efficient than one at 10 Gbits/sec, given that it moves four times as much data with the same channel spacing as today's 10-Gbit/sec product. But on 10-Gbit/sec products, the technology will allow channel spacing to go to 25 GHz soon and eventually to 12.5 GHz. This trend demonstrates that, while the game may end in a draw, a 10-Gbit/sec system will never be at a disadvantage when it comes to spectral efficiency.

There is little difference between grooming with a 40-Gbit/sec device or a 10-Gbit/sec device. For the same reason 10 Gbits/sec was valuable for carrying multiple 2.5 Gbits/sec, so too will 40 Gbits/sec eventually aggregate multiple 10-Gbit/sec signals.

Standard SONET ring protection re quires the protection be performed at the next higher multiplexing level. Today, 10-Gbit/sec rings are used to protect 2.5-Gbit/sec tributaries. With today's intelligent optical switches, protection can be provided in other ways, so there is no need to have a 40-Gbit/sec infrastructure to protect 10-Gbit/sec tributaries.

In the future, systems with 40-Gbit/sec routers will require 40-Gbit/sec technology. Today, however, technology exists to allow 40-Gbit/sec signals to travel over a 10-Gbit/sec infrastructure. This technology will protect existing investments in the 10-Gbit/sec infrastructure from obsolescence.

The real question involves defining the right line speed at which to run channels in the backbone. Generally, this decision should be based on the technology that provides the lowest cost per bit per mile. Multiplexing and switching technology can then adapt the traffic being carried to the line rates of the backbone.

No matter what carriers' hopes are for deployment of 40 Gbits/sec, they have to be able to test their implementation and continue to monitor and test the fiber pipes indefinitely. For 40-Gbit/sec deployments today, that will be an extremely expensive, if not impossible, task. Proven, economical test equipment isn't here yet.

The 40-Gbit/sec infrastructure is rather immature at the moment. The cost of an OC-192 tester is still in the $100,000 range, and few, if any, 40-Gbit/sec signal generators are on the market. At this stage, a single 40-Gbit/sec test unit, if it was available for purchase, will come with a half-million dollar price tag-not a pretty addition to the bottom line. And that sum, at best, would purchase a prototype unit.

On top of the cost of physical equipment, the cost of training and qualifying craftspeople to deploy 40 Gbits/sec, maintain 40 Gbits/sec, and know how to use new testers on 40 Gbits/sec will be significant. Technicians will have to become familiar with substantially more equipment, ranging from compensators to new 40-Gbit/sec testers.

Turning up 40-Gbit/sec links will re quire more detail, and link qualification will be more demanding than with to day's familiar 10-Gbit/sec systems. That will certainly be worth the effort when the time is right, but might not be worth the investment given current market conditions.

History repeats itself. In the early 1990s, the cost of training people and the cost of 10-Gbit/sec test equipment was excessive, links were more sensitive to transmission impairments, and people had to take care when deploying equipment. The incremental costs and training demands on a move to 40 Gbits/sec will be similar to those paid for in the move from 2.5 Gbits/sec to 10 Gbits/sec.

While 40-Gbit/sec transceivers offer the promise of a fourfold speed increase over 10-Gbit/sec networks, their promise of delivering it at a reasonable price is in the future.

Major chip vendors such as TRW's new telecom components company Velocium and Hitachi's OpNext are get ting into position to develop 40-Gbit/

sec technology using indium phosphide and gallium arsenide chip technology. TRW's Redondo Beach, CA, facility has rolled out the very first deliveries of these new components. TRW forecasts that higher-volume shipments are planned for mid-2002.

A few vertically integrated companies may be a bit ahead of that schedule, since they produce their own proprietary components, but there still is an 18-24-month gap between today's promise and tomorrow's reality.

It is still early for 40-Gbit/sec components. Eventually, 40 Gbits/sec will be the order of the day. But through 2002, volumes will remain small, no matter who produces the components, and that goes against good economics.

The time will come when 80 Gbits/sec and even higher speeds will exist. How ever, 40 Gbits/sec shouldn't be bypassed in anticipation of 80-Gbit/sec or 160-Gbit/sec systems. There is a chance the industry will leapfrog 80 Gbits/sec and go directly to 160 Gbits/sec, but that will take fantastic engineering.

In the meantime, developments in the 40-Gbit/sec market hold great promise. But looking beyond components, the entire 40-Gbit/sec infrastructure in general is still in its infancy. Even when 40-Gbit/sec technology be comes available for demonstration networks in the coming year, it will still be on the bleeding edge of fiber-optic technology. By 2003 or 2004, full successful deployments are expected, but that certainly is beyond any realistic expectation of widespread commercial installations in 2002.

For 2002, 10 Gbits/sec is still the answer. Down the road, by 2003 or 2004, that will change. Any manufacturer in business for the long term will provide its customers a smooth, compatible migration path from 10 Gbits/sec to 40 Gbits/sec. The two technologies will coexist in networks during the second half of this decade.

There are situations in the network of tomorrow where both technologies will be deployed to good advantage. Deploying only one or the other will cost service providers opportunity and efficiency in their networks.

There is also little doubt that the technological problems with 40 Gbits/sec will be solved over time. Although carriers had 10-Gbit/sec demonstration routes in the first half of the 1990s, it is only during the past three years that 10 Gbits/sec has been accepted, shipping in volume, for the long haul.

40 Gbits/sec versus 10 Gbits/sec is not an either/or decision. Over time, 10-Gbit/sec networks will migrate to 40 Gbits/sec, just as 2.5-Gbit/sec networks moved to 10 Gbits/sec.


Tom Mock is vice president of portfolio management at Ciena Corp. (Linthicum, MD). He can be reached by e-mail: tmock@ciena.com.

More in Market Research