In a recent post at EtherealMind.com, Greg Ferro examined possible implications associated with the impending dominance of merchant silicon in the networking industry.
Early in his post, Ferro reproduces a Broadcom graphic illustrating that the major switch vendors all employ Broadcom’s Trident chipset family in their gear. Vendors represented on the graphic include Cisco, Juniper, Dell, Arista, HP, IBM (BNT), and Alcatel-Lucent.
Custom switching ASICs haven’t gone the way of eight-track cartridges just yet, but the technology industry’s grim reaper is quickening his loping stride and approaching at a baleful gallop, scythe at the ready. Interrelated economic and technological factors have conspired, as they will, to put the custom ASIC on a terminal path.
There’s a chicken-and-egg debate as to whether economics occasioned and hastened this technological change or whether the causation was reversed, but, either way, the result will be the same. At some point, for switching purposes, it will become counterproductive and economically untenable to continue to design, develop, and incorporate custom ASICs into shipping products.
What’s more, the custom ASIC’s trip to the boneyard will be expedited, at least in part, by the symbiotic relationship that has developed between merchant silicon and software-defined networking (SDN).
Difficult Adjustment for Some
Commercially, of course, merchant silicon preceded SDNs by a number of years. Recently, however, the two have converged dynamically, so much so that, as Ferro acknowledges, future differentiation in networking will derive overwhelmingly from advances in software rather than from those in hardware. Vendors will offer identical hardware. They will compete on the basis of their software, including the applications and, yes, the management capabilities they bring to market.
For companies that have marketed and sold their products primarily on the basis of hardware speeds and feeds and associated features and benefits, the adjustment will be difficult. The bigger the ship, the harder it will be to turn.
There are some caveats, of course. While seemingly inevitable, this narrative could take some time to play out. Although the commercial success of merchant silicon was not contingent on the rise of software-defined networks, the continued ascent of the latter will accelerate and cement the dominance of the former. To the extent that the SDN movement — perhaps torn between OpenFlow and other mechanisms and protocols — fragments or is otherwise slowed in its progress, the life of the custom ASIC might be prolonged.
Timing the Enterprise Transition
Similarly, even if we presuppose that SDN technology and its ecosystem progress smoothly and steadily, SDN is likely to gain meaningful traction first with service providers and only later with enterprises. That said, the line demarcating enterprises and service providers will move and blur as applications and infrastructure migrate, in whole or in part, to the cloud. It’s anybody’s guess as to when and exactly how that transition will transform the enterprise-networking market, but we can see the outlines of change on the horizon.
Nothing ever plays out in the real world exactly as it does on paper, so I expect complications to spoil the prescience of the foregoing forecast.
Still, I know one thing for sure: As the SDN phenomenon eventually takes hold, the role of the switch will change, and that means the design of the switch will change. If the switch is destined to become a dumbed-down data-forwarding box, it doesn’t need a custom ASIC. Merchant silicon is more than up to that task.