All those predictions of dramatic increases in the use of network processor technology to develop communications infrastructure products have not come true. Despite the intuitive appeal of a technology that promised to combine programmability, ASIC-like performance and development savings, the expected explosion in network processor utilization has just not materialized.

Instead, the collapse of the communications equipment market forced customers to emphasize cost cutting over innovation and slowed the need to adopt network processors to enable new services. In addition, first-generation network processors failed to impress OEMs. Limited application performance and immature development tools and associated reference code resulted in learning curves, time-to-market and development costs similar to those of typical ASIC designs.

Today, network processors are being used to develop products for every level of the network infrastructure. Market demand and technology life cycles for communications equipment now favor net processor-based implementations of products and services. Increasingly, modular standards-based platforms such as those rooted in Advanced Telecom Computing Architecture (AdvancedTCA) reinforce the need for merchant silicon. Simultaneously, time-to-market requirements and availability of skilled technical talent continue to put pressure on controlling development costs. Fortunately, developers as well as vendors gained valuable insight from previous design efforts using network processors. Four key factors are combining to enable OEMs to reap significant benefits from network processor technology: improvements in second-generation silicon performance, training and experience with network processor software, greater opportunities for reuse of existing code, and better development tools.

Several network processor vendors now offer choices from OC-3 to OC-192/10 Gbits/second and beyond. The Network Processor Forum's industry-standard benchmarks let developers evaluate various architectures' performance characteristics. In addition, major network processor vendors are publishing performance data on key applications. For example, the Tolly Group, an independent testing organization, recently published results comparing the throughput of wireless gateways designed using Intel network processors with the performance of similar gateways based on other chip sets.

Developers have also become more savvy in their use of the processors. In general, initial experiences with network processors required OEMs to adapt to a new approach to product design as they made the transition from hardware-oriented development to an algorithm-based software methodology. Taking advantage of the flexibility of network processors places new emphasis on design methodologies such as structured software design to meet line rate performance and anticipate feature changes three to five years into the future. Software developers are now more knowledgeable about programming to specific network processor architectures and more proficient in using software development tools for those architectures. Development teams are applying key lessons from first-generation network processor designs to speed second-generation designs to market at reduced cost. As In-Stat's Eric Mantion recently noted, "Short time-to-market only comes in after you've gotten used to using software-programmable devices instead of hardwired ASICs. It is this potential cost savings that drives the need for NPUs to be used in so many places"( see "The Reincarnation of the Network Processor Market," In-Stat/MDR, December 2003).

In a comparison of NPU- with ASIC-based designs, one of our customers projects a 20 percent savings in development time and a 50 percent savings in head count over three generations of product design. As telecom equipment OEMs respond to carrier RFPs, network-processor-based products give the developer an advantage in responding rapidly to requests for variants and customizations.

The third factor favoring network processors is the availability of better applications software. As leading equipment vendors adopt the Internet Engineering Task Force's application programming interface for managing the control path and data path, a broader range of third-party software suppliers is developing applications software that can be easily integrated into the new platforms. Adoption of standards such as AdvancedTCA offers software vendors stable, open target platforms for development of emerging applications. In addition, leading network processor vendors are offering libraries of applications software to provide functionality at the required data rate. OEMs can select network processor architectures that enable them to integrate and customize the code to deliver highly differentiated services.

Greater reuse of software is also contributing significantly to lower development costs and decreased development risk. As an example, Intel has tracked software reuse across its own diverse network processor engineering teams. Data has been collected for more than 26 software building blocks ranging from MPLS/DiffServ, OC-48 AAL-5 and OC-192 POS components to edge router and resilient-packet-ring applications.

Intel's internal results to date across all building blocks indicate a median development savings of 53 percent (measured as person-days of effort required to find, analyze and integrate existing code blocks into an application compared with developing new code), with an associated 30 percent decrease in number of code defects (measured as preship defects tracked through system test) when compared with development of new code.

Fourth, network processor vendors and third parties continue to improve development tools to make programming easier with predictable performance. Cycle and data-accurate simulation and application performance modeling are widely available, helping developers estimate application performance, processing headroom and potential design bottlenecks before programming. Some advanced tools offer a packet-centric view of the network processor code that can accelerate exception handling, validation and verification.

Recently, Intel demonstrated new compiler technology that identifies available network processing resources and automatically partitions application components to meet developer performance specs. Such programming approaches make it easier to reuse code across applications and line rates.

These developments suggest that network processor adoption will not only continue but accelerate. Developers and vendors alike are getting better at utilizing net processors to speed new features and services to market. Communications equipment designers are deriving value from leveraging their experience and code base. This is leading larger OEMs with broad product portfolios to focus on a small number of net processor architectures that can be used broadly for multiple development projects. By working with a limited number of strategic suppliers, developers can realize the scalable performance, flexibility and development cost benefits promised by network processors.

Bill Mello is marketing-programs manager and Frank Schapfel is product line marketing manager at the Network Processor Division of Intel Corp. (Santa Clara, Calif.).

See related chart