Why a 400G Standard Might Draw Complaints
"I think we should have a flexible rate on the line side and also on the client side," said Ron Johnson, director of product management at Cisco Systems Inc. (Nasdaq: CSCO)'s Converged Optical and Routing Business Unit during a hallway conversation with Light Reading at Ethernet Expo Wednesday.
Johnson had just attended a panel session titled "Beyond 100G," where speakers noted that a 400Gbit/s call for interest -- a formal step preceding the start of standards work -- should be released early next year, based on work done by the Higher Speed Ethernet Consensus ad-hoc group. (John D'Ambrosia, chair of that group, confirmed in an email that he's working on a 400Gbit/s CFI but didn't specify the time frame.)
Based on the usual IEEE process, that means a standard would be finalized in late 2016 or in 2017, and that's what bugged Johnson. Given that demand for multi-100Gbit/s speeds has already started, albeit in extreme cases, 2016 is too long to wait for the next standard, he said.
"I just don't know how we can be innovative if we wait five years every time," Johnson says. "I just want to see it move faster, and it just seems there are a lot of things intentionally slowing us down."
His preference? A flexible model that would reflect the packetized nature of most traffic. The days of requiring certain speeds to suit time-division multiplexing (TDM) are passing, so why not let packetized traffic travel at more arbitrary speeds, maybe at multiples of some elemental speed, such as 25Gbit/s?
It's just an idea on paper, one that Cisco doesn't have working in the lab -- and one that not everyone in Cisco agrees with, Johnson conceded. But Cisco has been discussing the idea with customers and there's interest within Cisco in ultimately trying to standardize the idea, probably within the IEEE, he said.
This would be for the client side, the interface that points in the direction of the user. On the line side, a flexible-rate line standard could be a possibility soon. There's discussion in the International Telecommunication Union, Standardization Sector (ITU-T) about making OTU5 flexible, in a way similar to the way ODUFlex works, as panelists pointed out in a session about speeds beyond 100Gbit/s (the panel that fired Johnson's ire).
Such a standard would allow for traffic in 100Gbit/s multiples, and Johnson doesn't think that's flexible enough. "If a packet [flow] can justify a lambda or multiple lambdas, it should do that," Johnson says.
The panel included representatives from Fujitsu Network Communications Inc. and Huawei Technologies Co. Ltd. , both sounding enthusiastic about a 400Gbit/s Ethernet standard, and Infinera Corp. (Nasdaq: INFN), which can send 500Gbit/s superchannels with its latest gear.
Multiple modulation schemes would be desirable at 400Gbit/s, to allow for tradeoffs between spectral efficiency and reach. Fujitsu is hoping to get one transceiver that does two options. One would combine four channels of 100Gbit/s apiece into one 400Gbit/s superchannel using DP-QPSK modulation. This would have relatively long reach but would take up 150GHz of spectrum, said Randy Eisenach, Fujitsu's WDM and 100Gbit/s product planner.
The other option would be two 200Gbit/s channels using 16QAM modulation, taking up only 75GHz but sacrificing some distance.
An approach with modulation tradeoffs is probably best, and it would call for flexible-grid reconfigurable optical add/drop multiplexers (ROADMs), said Kent Jordan, senior manager of optical marketing at Huawei. "You could get a Swiss cheese effect over time when you get a fragmentation of the spectrum" and need to perform some spectrum defrag, he said.
Would the line side have to be standardized as well? Most carriers use the same vendor on both sides of an optical connection, so that line-side interface could stay proprietary, argued Vinay Rathore, senior director of solutions marketing at Infinera. It's not surprising he might say that, since Infinera can exceed 400Gbit/s on the line side.
Fujitsu's Eisenach didn't like the idea, though. "When you get down to the component level, you no longer have a standardized component or chip you could buy from a larger market, or have that wide, deep component or chip supply," he said.
— Craig Matsumoto, Managing Editor, Light Reading