Designing in 10Gbit/sec. network processors - Comment

Electronic News, May 6, 2002 by Jeff Cashen

FROM A PURELY PRAGmatic standpoint, any rational observer should be mystified by the recent flurry of questions concerning the industry's readiness for l0Gbits/sec. network processors. Perhaps the recent retrenchment in infrastructure spending has caused some people in the communications industry to completely lose their perspective. Maybe the simpler answer is that those that haven't yet figured out a viable 10Gbit network processor solution suffer under the mistaken belief that the technology isn't needed yet. Whatever the reason, it doesn't take much investigation to conclude that 10Gbits/sec. network processors have already become a vital factor for any systems company that wants to survive and thrive in both the near- and long-term.

Ultimately, real-world applications, traffic levels and users' requirements always drive the demand for higher bandwidth and faster network performance. No one would dispute that all three of these factors are relentlessly escalating and will continue to do so in the foreseeable future. The addition of millions of new Internet users each month along with the proliferation of broadband connections is exponentially multiplying the sheer amount of traffic being generated at the network edge. In parallel, convergence of data and voice traffic is multiplying the traffic loads and performance demands for metro and core network infrastructures.

High-speed network processing is becoming even more critical within increasingly heterogeneous environments that may have to support OC-192 and/or 10 Gigabit Ethernet traffic as well as provide flexibility for both legacy and new protocols, plus a variety of higher-level traffic management/engineering functions such as virtual private networks (VPNs). With floor space already a key limitation and huge expense for most providers, the flexibility to deploy OC-192 line cards that can then be channelized down into OC-48 increments can allow for a much more advantageous cost structure.

For system providers, the reality is that 10Gbits/sec. requirements are already a fundamental part of virtually every RFQ being issued by tier-one and tier-two service providers. It is now impossible for a system vendor to effectively participate across multiple market segments without having a credible 10Gbits/sec. solution within its family of product offerings.

From a system developer's standpoint, the most immediately pressing issue is the need to make fundamental design decisions today that will smoothly accommodate the transition to l0Gbits/sec. networks, without entailing a costly disruption between incompatible architectures and/or the risk of falling behind against the competition. Given the typical 18 to 20 months required to bring new system designs to market, any vendor that doesn't have 10Gbit/sec. developments moving through the pipeline now has already fallen well behind the competitive curve.

In today's world, the key challenges have shifted more toward effectively managing the migration path, reusing development investments and shortening overall time-to-market cycles. In such an environment, systems manufacturers are finding they no longer can afford the luxury of developing custom ASIC-based systems from the ground up. As performance requirements move to l0Gbits/sec. and beyond, most forward-looking system developers have turned to proven merchant network processor solutions that provide a reusable and extensible technology base, within which development investments can be smoothly leveraged from one generation into the next. By designing OC-48 cards using established NPU families that include software-compatible l0Gbits/sec. offerings, these system developers have simultaneously solved the l0Gbit issue to the satisfaction of their end-customers, have minimized their development costs and have shortened their time-to-market.

The bottom line is not so much to be right about exactly when the market for l0Gbits/sec. is going to fully blossom, but rather, to be ready to take maximum advantage when it does. Tomorrow's market leaders have already sized up the situation and invested in the design decisions that will keep them at the head of the pack for l0Gbits/sec. and beyond.

Jeff Cashen is the VP and GM of the switching and network processing division of Applied Micro Circuits Corp. in San Diego.