Getting on the Same Wavelength

In a classic case of one hand not knowing what the other is doing, telecom’s technocrats have created two incompatible standards for 10-gigabit Ethernet. One (802.3ae-2002) was developed by the IEEE for the LAN and clocks in at 10.3gbps. The other (G.709) was developed by the ITU-T for the WAN and is 9.9gbps, which is equal to OC192.

That’s a problem because 10gigE LAN PHY is common in core and edge routers used to deliver Ethernet services, making it necessary for optical networks also to handle this protocol.

Not surprisingly, there are a number of methods emerging for dealing with this discrepancy — several of which have been submitted to the ITU-T for incorporation into the optical transport network (OTN) standard. The ITU-T Study Group 15 is expected to consider these proposals at its October meeting in Geneva, Switzerland.

One of the most common approaches is called “overclocking.” While the 10gigE WAN PHY standard — with 9.9gbps plus 7 percent overhead — purposefully maps neatly into OTN standard OTU2 10.7gbps frames, the 10gigE LAN PHY does not. A data rate of 10.3gbps plus 7 percent overhead equals 11.1gbps. However, this overclocking method is popular. “The good thing about it is it’s simple,” says Ram Orenstein, director of product line management of xWDM for ECI Telecom Ltd. He explains that it’s transparent — what goes in comes out. “The downside is it’s not the standard OTU2 as in the G.709 standard, so you don’t get all the counters for performance monitoring and tandem connection monitoring,” which is important in tracking waves across multivendor domains.

A second shortcoming, says Orenstein, is that nonstandard frames preclude the use of OTN-based optical cross-connects to switch wavelengths and subwavelengths. And, he adds, operators cannot use standard regenerators in a long-haul or mesh network.

The problems with overclocking multiply when you raise the data rate up to 40gbps. “Four OTU2s (at 10.7gbps each) can fit into an OTU3 without a problem,” says Vinay Rathore, director of marketing for Ciena Corp., explaining that an OTU3 is designed to carry an OC768 (39.812gbps) transparently. However, if you have overclocked the OTU2, then you have to overclock the OTU3, which can test the timing tolerances of a client like Ethernet, he says.

Rathore says 45gbps seems to be the limit for OTU3. “Going any further can cause more issues — not just in terms of being able to fit it, but in being able to carry it,” he says, citing the limitations of off-the-shelf components. On operator request, Ciena successfully completed interoperability testing on an overclocked 11.05gbps Ethernet connection across a multivendor network and multiplexed into an overclocked OTU3, Rathore says.

Rather than overclock, vendors also have proposed ways to preserve OTN functionality. One way, called GFP mapping, is simply to drop the Ethernet overhead part of the payload altogether. “This way you reduce the actual bandwidth to the size that fits the OTN framing, so it’s standard,” says ECI’s Orenstein. “The problem is that by removing the Ethernet overhead bytes, you lose some functionality of the Ethernet layer.”

Click to Enlarge One method for mapping 10gigE LAN PHY to WAN PHY using OTN framing is to carry the excess client payload in the unused 7 bytes of OTU overhead, shown in this diagram in yellow.

This is particularly true of networks using gear from Cisco Systems Inc., which makes proprietary use of the overhead bytes for OAM functions. An improvement on this approach is to keep the Ethernet overhead and put it into idle packets for extraction at the other end, says Orenstein.

This idea was made public in a January 2006 proposal from silicon vendors Intel Corp. and Fujitsu Microelectronics Europe. The two are working with Lucent Technologies Inc. and ASIC designer AimCom BV to develop a device called McPOM, or multichannel packet OTN mapper. McPOM will be incorporated into Lucent’s LambdaXtreme Transport DWDM system, says Matthias Berger, head of high-speed optics development within Lucent’s Optical Networking Group. He declined to offer a delivery date, noting that the trial silicon only recently has become available. The Intel-backed solution removes the 66b/64b encoding (described in 802.3ae, clause 49), and looks for idle packets to downclock the signal to 9.9 or 10gbps and recodes, says Berger.

ECI’s Orenstein says this solution is a good one, but compromises throughput. “I will be consuming some of it for my internal mapping problem,” he says. “I know that Intel says it’s only 3 percent, but we think it’s more complex than that.”

Ciena, ECI and others are taking yet another tack by mapping the Ethernet overhead into unused bytes in the OTN frame (see diagram above). “It’s enough to make sure that all the Ethernet trail is mapped into the OTN container with 100 percent throughput,” says Orenstein. “Then, of course, we would extract these bytes from the OTN overheard and place them back at the 10gig LAN port.”

This idea has been proposed to the ITU-T by chipmaker Applied Micro Circuits Corp. (AMCC). Ciena, which makes its own chipset, also has floated a similar proposal with the standards body, Rathore says.

The problem with this method, says Lucent’s Berger, is layer violations. “That means they are going from the client layer to the G.709 layer to transport the data,” he says, explaining that if G.709 ever makes use of those header bytes, the mapping scheme could be rendered incompatible. “For the time being, it’s OK, but it might generate issues in the future. That’s why Intel came up with this other mode.”

Ciena’s Rathore concedes there is a downside to using the header bytes if you are crossing multivendor networks since the gear has to strip off the information in the same way. “To be honest, the implementation is as simple as telling your data path where you want to put the payload,” he says. “It’s not rocket science.”