XFP Module Gets a Shrink
It's being called SFP+, and early standards talks are happening under the auspices of the American National Standards Institute (ANSI) Technical Committee T11 -- the group presiding over Fibre Channel. The most recent meeting was held in San Jose, Calif., hosted by Broadcom Corp. (Nasdaq: BRCM), and the next session will likely come this month during the T11 interim meetings.
More than 30 companies reportedly were represented at last month's meeting -- mostly systems vendors, but also some module makers such as JDSU (Nasdaq: JDSU; Toronto: JDU) and cable/connector firms including Molex Inc. (Nasdaq: MOLX) "A lot of the data center people are interested," says Bob Salem, a Broadcom senior product line manager.
SFP+ would be a smaller module than XFP. It removes electronics such as a chip for clock and data recovery, putting them on the linecard instead. This lets the module shrink to the size of the small-form pluggable (SFP) used for 2.5-Gbit/s ports.
Why does that matter? For starters, the engineering is easier when things are added to a linecard, as opposed to the module. A shift to SFP+ could lower the cost of optical modules. "A lot of our systems customers are telling us they need to find ways to lower the cost of a link. This is one way," Salem says.
Of course, SFP+ opens up a problem, too, because the linecard now has to carry one more chip in addition to an Ethernet physical-layer chip, media access controller (MAC), and others. "On the back end, it still takes up a lot of real estate" on the linecard, says Bill Ryan, a product marketing manager with Foundry Networks Inc. (Nasdaq: FDRY).
The smaller size of SFP+ could also boost linecard density to at least 24 ports per card from 16 with XFP, says Paul Voois, CEO of chipmaker ClariPhy Communications Inc. But wait, there's more: Putting the electronics outside the module also opens the possibility of serving multiple lines per module, the way Gigabit Ethernet does. "That's how you really drive down costs," Voois says.
Assuming SFP+ gets used for 10-Gbit/s Ethernet as well, it would be the latest in a series of module shrinks. The industry started with the 300-pin multisource agreement (MSA), which is being replaced by the smaller Xenpak and X2 module types, which in turn could be replaced by the even smaller XFP.
If SFP+ is such a great idea, why wasn't it done before? The problem was that a 10-Gbit/s signal got garbled when traversing an ordinary linecard. Xenpak, X2, and the 300-pin MSA correct this by splitting a signal into multiple parallel lanes. XFP retains a serial 10-Gbit/s stream but is usually followed by a serializer/deserializer (SerDes) on the linecard that does the splitting into parallel streams.
In the last few years, signal-processing tricks from the likes of Broadcom, Marvell Technology Group Ltd. (Nasdaq: MRVL), and Rambus Inc. (Nasdaq: RMBS) have overcome those problems. "A few years ago, people said you can't have 10 Gbit/s on the PCB [printed circuit board]. Now that's not much of an issue," Salem says.
SFP+ gets its name from being the same physical dimensions as a Small-Form Pluggable (SFP) module, the standard used for many ports at 2.5 Gbit/s and slower. That brings up the question of whether the SFP+ standard should be designed so that an older SFP module could work in the same slot. If the answer is yes, then equipment vendors could outfit systems with ports ranging from 1 Gbit/s to 10 Gbit/s.
Prospects like that have the Ethernet crowd muscling in on the standard. "The Ethernet guys want this to do Ethernet as well, and the Fibre Channel group are sort of like, 'Why should we bother?' That doesn't bother the Ethernet group. They just come in and disrupt the meetings anyway," Voois says.
(Separate from all this, there's the question of whether 8-Gbit/s or even 4-Gbit/s Fibre Channel is all that urgent -- see High-Speed Drivers.)
Some module makers say they want to sample SFP+ products by the end of this year, or early in 2007. As for the systems vendors, they're taking the usual detached view, saying there's no urgency to this yet.
"We're evaluating SFP+ for the long term," Foundry's Ryan says. "We're satisfied with the density of XFP; the biggest issue is the cost of the optics."
— Craig Matsumoto, Senior Editor, Light Reading