Gigabit and 10-gig Ethernet fiber SFP slots and lock-in: Don't buy Cisco, Extreme, 3Com, HP, or Intel gear

Ethernet has been such a vast success in the computer networking market that it is starting to make serious inroads into the fiber optic data communication market. That market had previously been dominated by telco standards such as SONET, and some niches like Fibre Channel, FDDI, and ESCON. Ethernet's broad acceptance and its low cost and its rapid evolution and its adherence to strict compatibility standards have interacted to produce a totally dominant standard in computer networking. This dominant position is held by a technology that everyone is free to implement and to innovate on.

What does this have to do with fiber optics? Ethernet standards for fiber optic media have been published for decades. The earliest communicated at 10 megabits per second, like the wired Ethernet of its time, but it was expensive and saw only niche use. Fast Ethernet (100Base-T) on "Cat 5" copper wires is the kind of Ethernet that most people have seen. Fiber standards for that speed are also available (100Base-FX), and were also used in niches, though FDDI got there first and saw more use at on fiber at 100 megabits than 100Base-FX.

A major innovation in Fast Ethernet opened the floodgates. Full-duplex Ethernet eliminates all restrictions on the total distance that an Ethernet network can cover. Old Ethernet required that the network be smaller than a few thousand meters, so that packet collisions could be reliably detected. There are no collisions in full-duplex Ethernet. Distances are still limited, but only by the physical properties of the medium, such as twisted pairs or fibers, and these can be extended by adding switches (which also act as connection points). Once the distance limits of the old Ethernet protocol were eliminated, it became possible to use Ethernet at the farthest limits of fiber optic technologies, which can send signals reliably for many dozens of kilometers. Suddenly, cheap and compatible Ethernet gear could be used to build networks the size of a city, or larger.

The creation of Gigabit Ethernet (by Granite Systems and others) made Ethernet not just an available technology for wide area networks, but a compelling one. The only things that ran at similar speeds, like SONET gear, cost upwards of US$30,000 per node and are full of crazy complexities. Gigabit Ethernet gear can now be had less than US$500 per node, and it just plugs in to your existing Ethernet network. And 10-gigabit Ethernet gear is also readily available, with 40-gigabit gear available at the high end.

Gigabit Ethernet can use a variety of kinds of fibers, of widely varying cost, so gigabit fiber equipment is designed around an industry standard called "SFP" (Small Form Factor Pluggable) modules, which are about the size of a pack of gum. The standard is defined by a published "SFP Multi-Source Agreement" that specifies the physical dimensions, connectors, and signalling so that many vendors can build compatible products. Products such as routers or network interface cards provide one or more SFP slots, into which the user can plug an SFP module appropriate for that connection. The fiber itself plugs into a connector on the SFP module. Many dozens of kinds of modules are available, for use with different types of fiber, or for different wavelengths within a fiber, or for communication over shorter or longer distances. As with all Ethernet gear, the customer can mix and match components to meet their needs and desires as they build their network, and add new pieces from different vendors as their network expands.

Well, almost. Some vendors are trying to prevent their customers from making their own choices in the Ethernet market.

Cisco, the leading vendor of Ethernet switching and routing gear, has slipped a proprietary gotcha into the fiber optic slots in their switches. Cisco equipment won't accept "just any vendor's" compatible SFPs. The firmware in the switch checks the identifying data in the SFP's internal Flash memory, and if it isn't built by Cisco, the switch refuses to enable that slot. This practice has been taken up by some other vendors as well. Other vendors do not do it.

The obvious way to stop this practice is for customers to refuse to buy equipment that does it. Of course, the vendors do not advertise it - the customer only finds out when they discover that their "industry standard" component is deliberately refusing to talk to another "industry-standard" component. By that time, it's often too late to throw the entire thing back at the vendor and find a different vendor. This disreputable lock-in technique has only seen limited discussion on the 'net. Thus this article.

Good Vendors whose products DO NOT enforce lock-in on SFP slots

(I will be happy to add entries to these tables, as I receive confirmation from personal experience, from users and/or from vendors. Please send email.) It's getting a little tiresome to keep hearing that a company builds a product that supports open standards, but if the customer actually plugs in something compatible with that open standard, then the company won't support that customer. Do these companies support their product if you plug them into another Ethernet-based product, such as a computer? Or are their routers and switches only supported when they aren't plugged into any networks? Why is an SFP slot any different than an Ethernet cable slot?

Bad Vendors whose products DO enforce lock-in on SFP slots

As fiber-optic Ethernet grows out of its niche into a huge market, customers can keep Ethernet a fully "plug-compatible" system if they exercise a little care in their choice of vendors. Be sure to design your network so that gear which deliberately violates compatibility is excluded from your network. The people who have to maintain and improve your network in the future will thank you -- and you'll be standing in solidarity with every other intelligent customer.

Up to John Gilmore's Sysadmin index ... Up to John Gilmore's home page