Innovation and Merchant Silicon: Not An Oxymoron

The rise of merchant silicon in network switches and routers can have two potential outcomes. The first is a race to the bottom on price and quality.

Today, networking vendors enjoy large profit margins, and many manufacturers see a market that is ripe for flooding with low-cost and low-quality products. A 20% profit margin on a merchant silicon-based switch is a great return for a manufacturer used to the razor-thin margins on desktops and servers.

The other outcome is that merchant silicon frees up vendors to develop new features. Instead of funding a silicon development project that may take three to five years to come to market, networking vendors can work on cheaper SFP optics, better power supplies or more reliable code.

We may even see funding for next-generation features such as 40 GbE and 100 GbE, OpenFlow and OVSDB support, and wider support for newer protocols such as LISP.

One example of the second outcome is the Broadcom Trident2+ chip for Ethernet switches. The Broadcom Trident2+ has two key form factors: one with 32 x 40-GbE Ethernet ports presented as QSFP and the other with 48 x 10-GbE SFP with six 40-GbE QSFP ports. The Trident2 is shipping in multiple products including the Cisco Nexus 3100, the Dell S6000, the Nuage Networks VSG 7850, and the forthcoming S68 from Pluribus Networks.

[Cumulus Networks separates the network OS from switch hardware. Get insight into the potential impact of this strategy in "Cumulus Brings the Red Hat Model to Networking." ]

While all these vendors use the same chip, they bring distinct offerings to the market. Cisco emphasizes the OpenFlow compatibility of the Nexus 3100. The Dell S6000 operates at room temperature and is partnering with VMware NSX for VTEP support. Pluribus offers a low-latency switch with an open operating system based on OpenSolaris. Nuage Networks can now offer VTEP integration for its Virtualized Services Platform.

Low-Cost ECMP Networks

All of these products are well-suited to Equal Cost MultiPath (ECMP) Ethernet fabrics that use 1 RU switches to build large, high-performance Ethernet networks with a leaf/spine architecture.

(click image for larger view)

There's a lot of Ethernet in a 32-port 40-GbE switch when customers stretch their budgets a little. A 40-GbE QSFP interface can break out to a 4 x 10-GbE interface with a suitable adapter. It can become a leaf switch with a configuration of 28 x 40-GbE ports with QSFP breakouts for 112 10-GbE ports for servers with 4 x 40-GbE reserved for uplinks, for a 7-to-1 uplink contention. Or consider a more practical mix of 24 x 40 GbE as 10 GbE with 8 x 40 GbE for uplinks, for a more practical 3-to-1 uplink contention.

I do believe we'll see the cost of network hardware go down, but not to the degree that a switch or a router becomes a commodity like a server.

That's because merchant silicon offers vendors to opportunity to bring new features to market quickly and with limited risk; if those features have value to customers, customers will pay a little more for them.

And even as hardware costs decrease, network spending will remain the same because customers will refresh their equipment more frequently.

The market is already stocked with cheap networking gear. If cheap is a winning strategy, why hasn't Netgear, D-Link or even Huawei achieved greater success in the last five years?