Sizing Up Xenpak
In the press release, Ed Cornejo, Opnext’s director of applications marketing, is quoted as saying: “Xenpak has broad industry support and is generating a lot of interest from our telecom, datacom and storage customers.” (See OpNext Intros 10-Gig Modules.)
True, or false?
There’s little doubt that component vendors are giving Xenpak their backing. A total of 24 of them have now signed the agreement, which ensures that system vendors will get a wide choice of module suppliers (see Agilent, Agere Drive 10-Gig Ethernet ). At least five of the module makers have already announced products. For the record, they’re Opnext, Agere Systems (NYSE: AGR), Agilent Technologies Inc. (NYSE: A), Blaze Network Products Inc., and Optillion AB.
But are system vendors actually going to buy Xenpak modules? The jury’s still out on that question.
For starters, 10-gigabit Ethernet is in its early days. But, assuming that it does take off, Xenpak could come up against issues with the size and "feel" of the packaging, according to industry watchers.
Drew Lanza, general partner at Morgenthaler doesn't mince his words. "Xenpak isn't going to work," he says. "It's too big."
Systems vendors need to drive costs down, and they need higher density optics to enable them to do this. But Xenpak, Lanza contends, doesn't go far enough in this respect. At 4.36 inches long by 1.4 inches wide and 0.46 inches high, the Xenpak package is actually smaller than a standard 300-pin OC192 Sonet transceiver, but it's still considerably larger than the typical packages that datacom companies use for Ethernet.
And the fact that it's something "in between" is another irritation. Systems vendors being the conservatives they are, Lanza figures that neither datacom guys nor Sonet guys will be comfortable with the new packaging.
But Agere, which put the original proposal together with Agilent, says that it was responding to customers' needs when it drew up the Xenpak design. "The first thing we did was shop around our data customers," says Julie Eng, a director of product development at Agere. "At the time they viewed it as fine. But since then some of them have come back and said they need something smaller, and we're looking into that."
Agere says that one of the key things vendors asked for was to make the module hot pluggable, so system cards could be stored without optics and populated at the last moment before shipping to a customer. But Agere went about this in a strange way, in Lanza's view, with a design that requires the systems vendor to cut a hole in the printed circuit board and then slide the Xenpak module in.
Agere says there is method in its madness. "The primary reason we did this was so you can flow air over the top and bottom of the module, so you would get a little break on the thermal problem," Eng explains.
But Lanza faults this logic. "Density should be dictated by form factor, not by heat or connectors," he says. "What we're seeing is that you don't need that much space to dissipate heat. The systems guys don't like it."
So can Xenpak be made smaller? Agere's Eng says the company is working on it. But, at the end of the day, it may come down to the fact that there is a lot of functionality and electronics inside the Xenpak package. And if people want to start using enterprise-style optics, such as the small form-factor pluggable (SFP), for 10-gigabit Ethernet, then some of the functionality will have to come out of the package and go back onto the board.
"Why is a Cadillac bigger than a mini?" says Lanza. "It's got more stuff crammed inside."
— Pauline Rigby, Senior Editor, Light Reading
http://www.lightreading.com
Too Big - or Too Small? We have see this played out before. First implimentations are big, later generations trend to small. The extra room afforded in a roomy package allows vendors to differentiate. Long reach, XLR, ITU lambda, even MEM/VCSEL Agile DWDM ready may become possible some day in the envelope if the RF and eyesafe topics can be managed.
Front Pluggable? Vendors will have a hard time manageing that. What prevents a customer buying a cheapo 1310 nm SR blade and popping in a 1550 nm XLR ITU grid version later on the QT? Where does that leave the vendor's FDA laser safety, ALS, EMI and heat story? Containing RF emissions on a terabit system with up to 8 XENPAKs per blade may prove a real feat with the double whammy of X band front access gasketting plus containing the laser's RF launch out the apreture.
Interfacing rates? Nextgen CMOS PHYs and MACs will likely have 10G CMOS I/Os that can interface directly to serial transceivers with integral CDRs.
Cutting board slots? Biggest deal is the manufacturing process for the socket. What kind of IR reflow and test implications arise. Routing out the slot is nothing to be afraid of. But cooling is a big deal. At 10G you have a lot of impact if you lose your link. Operating the laser junction at 3-5C lower temperature during its life can dramatically extend MTBF so its worth it. Especially if it helps you avoid the need for exotic 80 or 85C case limits.
In summary I say look beyond the package into the materials and system implications. There will be 10G materials for trunking and LANs and dense, dependable 10G materials for the big cores. One product won't likely fit all needs.