Merchant OC-192 silicon represents next wave of 'disruptive technologies'

SPECIAL REPORTS / Optical Networking/WDM

Off-the-shelf components can speed time-to-market and increase functional flexibility.

BIDYUT PARRUCK, Azanda Network Devices

If you wanted to build OC-192 (10-Gbit/sec) communications systems a year ago, you would need to design ASICs for most key functions in the system. But that's about to change. By the end of 2001, merchant silicon will be available for all functions needed to build OC-192 equipment. By the end of 2002, OC-192 performance will become available on all types of equipment in the network.

The widespread availability of OC-192 standard silicon products will have a profound impact on communications. As with any "disruptive" technology change, system manufacturers, carriers, and end users should revisit their strategies to ensure they account for the forthcoming widespread availability of OC-192 products and 10-Gbit/sec bandwidth.

The impact of merchant market OC-192 silicon is best appreciated by first reviewing the traditional approach of using in-house ASICs to create OC-192 systems. Only the largest system houses could afford this approach.

As shown in Figure 1, there are multiple functions in a system that must support OC-192 speeds, including traffic managers, packet classifiers, switch fabrics, framers, and physical layers (PHYs). Many of these functions require multiple chips. Developing ASICs for these functions costs tens of millions of dollars, development cycles for these ASIC chips are at least 18 months, and until each chip functions properly, the system cannot be built.

Figure 1. A number of different chips may be necessary to process OC-192 signals. While chips for these line rates previously had to be developed in-house, merchant silicon is now available that can significantly accelerate design cycles.

Recent trends have made in-house ASICs even more difficult to develop. The trend toward multiprotocol silicon makes the verification of ASICs much more difficult, since there are more configuration options and traffic combinations that need to be tested during the design phase. A further complication for ASIC developers is that the Optical Internetworking Forum (OIF) and Network Processing Forum (NPF) have defined standards for interconnecting networking chips. These standards benefit merchant silicon providers and their customers by allowing chips from different vendors to be mixed in a given system. But for ASIC developers, the availability of standards increases development costs, because a standards-compliant interface must be designed and verified if the ASIC is to be able to interoperate with existing and future merchant silicon.

Finally, from an overall industry standpoint, it is inefficient to have each system house separately make investments in OC-192 ASICs-there is not enough semiconductor design talent or funding available to justify such duplication of resources. For all the above reasons, the ASIC approach has been a route of last resort, used only because of the absence of merchant OC-192 silicon.

For large system companies, choosing merchant OC-192 silicon over ASICs provides faster development cycles, lower nonrecurring engineering costs, lower per-unit costs, and lower risk compared to in-house ASICs. For small system companies, merchant silicon is the only route affordable, particularly given the current limitations of new venture funding and public market funding. For these reasons, the switch from in-house ASICs to merchant silicon is expected to be extremely rapid.

Semiconductor companies can only afford to develop standard products for high-volume markets. Fortunately, the prospects for the OC-192 market are excellent.

The infrastructure for widespread OC-192 adoption was created by the over-installation of fiber during the late 1990s Internet gold rush. The capacity of this fiber continues to expand due to the ever-increasing number of wavelengths carried on DWDM systems.

Figure 2 shows that the exponentially increasing demand for more bandwidth for data traffic will continue. Traditional voice traffic, with its slow growth, has become a smaller percentage of the overall traffic. The demand for more data bandwidth is being stimulated by faster last-mile technologies such as VDSL, passive optical networks (PONs), and 3G cellular. Enterprise-to-enterprise connections are also demanding more data bandwidth, fueled by applications such as XML, B2B interactions, enterprise applications integration, supply chain management, and Web-based initiatives like Microsoft's .NET.

Figure 2. Traffic patterns have changed significantly over the past few years. Data promises to become the overwhelmingly dominant traffic type in carrier networks.

With OC-192c silicon, the entire width of the pipe can be used to burst data between locations. This type of bursty, ultra-high-speed traffic allows new types of applications such as connecting two server rooms, accessing remote storage farms, or implementing advanced content distribution networks.

Architectural and interconnection standards are also driving demand. Over the last few years, the key functions of the line card have separated into well-defined functions. These include the traffic manager, packet classifier, switch fabric, framer, and PHY. There are now specialized semiconductor companies focusing on each of these functions. The OIF and NPF interface standards allow mixing and matching of chips from different vendors to best meet the requirement for different types of equipment. Standard architectures and interfaces make it much easier to add OC-192 speeds to any type of equipment and will accelerate the implementation of OC-192 speeds.

As the optical core of the network continues to mature, electrical intelligence migrates to the edge and metro systems. If the core is optical, then all traffic types must be combined before being sent over the fiber. Accordingly, a frequent requirement for all types of access and edge devices is the ability to intelligently handle multiprotocol data and provide different qualities of service for different types of traffic. With multiprotocol merchant silicon, it is possible to develop a single line card that handles all traffic types: packet, cell, and legacy TDM.

To address the broadest range of equipment types and the largest possible markets, silicon vendors are packing a wide range of features into their chips. For example, OC-192/OC-192c multiservice traffic management technology is now on the market that supports one million flows, multiple quality-of-service levels, support for MPLS and ATM, plus advanced statistics that enable billing for enhanced services. That opens up a wide range of new equipment types that can be quickly designed at OC-192 speeds.

There are a number of ways system designers can exploit merchant market OC-192 silicon to create competitive advantages:

System designers should plan their equipment's capabilities around best-of-breed merchant functions. That requires changes in the ways products are defined. When ASICs are used to build products, marketing and engineering can agree to any reasonable wish-list of features. But if merchant silicon is to be used, the menu of capabilities at the system level must be guided by the features available in the merchant silicon.
System companies need to establish close working relationships with the leading best-of-breed merchant silicon companies. By working closely with its semiconductor partners, a system company can guide the features implemented in the next generation of silicon to ensure the necessary foundation is laid for its next generation of equipment.
System designers should look to software to create product differentiation. While some system companies may argue that ASICs provide an important source of differentiation, it is unlikely that in-house ASICs, with their long development cycles, high costs, and all-or-nothing risk profiles, can successfully compete over the long term against the faster-time-to-market, lower-cost development profile of feature-rich merchant silicon. Accordingly, software becomes a key source of differentiation.
System companies need to prepare for much faster time-to-market cycles. ASIC-based systems typically take two or more years to develop due to the long cycle time of ASIC development. By using merchant silicon, system companies can cut development cycles to less than one year. In many cases, software design and test cycles will become the limiting factor.
Points 3 and 4-product differentiation through software and elimination of software development bottlenecks-both argue that system companies should place renewed emphasis on software as a distinctive competence. The increased resources that system companies direct toward software engineering can be funded from the savings arising from reducing or eliminating ASIC development.
Participate in and leverage the work done by standards organizations. The OIF and NPF have defined electrical and software interfaces that allow a building-block approach to the construction of OC-192 systems. Steer clear of semiconductor vendors pushing proprietary interfaces-the intent of these suppliers is to lock the system vendor into one-size-fits-all solutions. Leverage of merchant silicon is best accomplished by making best-of-breed component selections for each function in the system and insisting on standards-based interfaces from vendors to promote this mixing and matching.
Design for upgradability and to create futureproof equipment. Design OC-192 systems so they can be upgraded to OC-768 simply through a line-card swap in the field. Select silicon vendors that offer scalable architectures that provide a smooth migration path from OC-192 to OC-768 and beyond. Design equipment with multiple protocols-carriers want to dynamically change the mix of protocols handled by equipment. By using multiprotocol merchant silicon to support all traffic types and building equipment for field upgradability, system designers can extend their time-in-market, while generating future revenue from line-card upgrades.

Carriers also can take a number of steps to create competitive advantages from merchant market OC-192 silicon:

Carriers should streamline their field-trial process. Merchant market silicon will create a tidal wave of different types of OC-192 equipment, each of which can open new markets. In addition, the shift in performance to OC-192 means carriers can create more bandwidth with existing installed fiber plant, thereby generating more revenue from each fiber. However, none of these benefits can occur until the new equipment goes through field trials. Accordingly, fast field-trial processes can be used by carriers as a distinctive competitive advantage.
Carriers should keep their network architectures flexible to rapidly respond to changing demand patterns. That includes moving intelligence from the core to the edge and selecting equipment that supports advanced capabilities such as MPLS, network-based virtual private networks, and virtual routing. Carriers should also look for equipment that can be remotely provisioned by software control.
Carriers should demand multiprotocol equipment and line cards from their vendors. With multiprotocol systems and flexible system architectures, carriers can dynamically change the mix of traffic at any time. Carriers can provision new bandwidth faster and are more capable of winning contracts for high-margin, short-term special events. In addition, multiprotocol equipment futureproofs carrier investments by eliminating the need to perform "forklift upgrades" when demand patterns change.
Carriers should extend the life of existing equipment by asking vendors for multiprotocol OC-192 line-card upgrades for existing chassis that are already deployed in the field. That's a quick way to expand system capabilities under today's tight capital-expenditure budgets.
Carriers should plan for a smooth migration to OC-768. New fiber plant should be selected for OC-768 DWDM compatibility. System vendors should be encouraged to show clear "forklift-free" migration paths from OC-192 to OC-768.
Carriers should encourage their system vendors to use standards-based protocols, not only on external interfaces, but also at the chip-to-chip level. Systems built with standard chip interfaces are more easily upgraded by both the system vendor and carrier. By using chips with standard interfaces, a system vendor can quickly redesign part of a line card to add new capabilities. For carriers, the ability to perform upgrades by swapping line cards is a key factor in extending equipment lifecycles.

The widespread availability of OC-192 technology will further increase the percentage of the traffic mix that is not traditional voice. While traditional voice will continue to be an important revenue stream, carriers increasingly will need to look to nontraditional services for future revenues.
Carriers should select equipment with sophisticated traffic engineering that has support for easily creating enhanced revenue sources. Fine-grained bandwidth control and multiple quality-of-service levels can be used to create multiple tiers of service. For example, these capabilities could be used to create Ethernet WAN services with guaranteed bandwidth and fair usage of excess bandwidth. Equipment with advanced statistics gathering and per-flow granularity can be used to deploy new billing approaches. The latest OC-192 merchant silicon makes that feasible by allowing carriers to manage and gather statistics on up to one million separate flows in a single OC-192 link. For example, a carrier could offer leased-line replacements, with the ability to support virtual connections between customer sites with specific service-level agreements (SLAs) and billing based on the SLA terms. Carriers should opt for equipment that supports both OC-192 and OC-192c. OC-192c allows traffic bursts for a single subscriber that can utilize the full 10 Gbits/sec of bandwidth.
Carriers should select equipment that provides streamlined network monitoring to reduce maintenance costs. In many cases, the same statistics gathered for billing by OC-192 silicon can be used for maintenance and network monitoring.

The PC revolution moved from 50 MHz to 1 GHz over about a 10-year period. The bandwidth revolution is moving much faster, with far-reaching effects that even the best crystal ball has trouble predicting.

However, the overall conclusion is clear: End users should plan now for lots of inexpensive bandwidth. Capabilities that are impractical due to bandwidth restrictions today will become commonplace in the next few years.

Not only is this new bandwidth low-cost, it is also highly flexible. It is capable of supporting multiple protocols and it can be rapidly provisioned, allowing its use for one-time special events. It can support real-time, high-resolution video and audio, allowing it to be used as an alternative to traditional broadcast technologies (TV and radio).

To create competitive advantage from this new bandwidth, end users should consider both improvements to existing business processes and entirely new "out of the box" alternatives. System vendors and carriers should be considered partners in exploring what might be feasible with more bandwidth.

Bidyut Parruck is a founder and chief technology officer at Azanda Network Devices (Sunnyvale, CA).