February 16, 2018    

The IEEE has launched a Study Group to explore fiber-efficient ways to implement 200 Gigabit Ethernet and 400 Gigabit Ethernet via multimode fiber. The Study Group's existence could be seen as an acknowledgement that current multimode fiber PMDs for such transmission rates (400 Gigabit Ethernet, as the 200 Gigabit Ethernet specifications are still being developed) may prove a bit unwieldy to deploy. But it also demonstrates that data center operators foresee a continuing need for multimode fiber options at data rates beyond 100 Gbps.

The multimode fiber PMD for 200 Gigabit Ethernet is in the hands of the IEEE P802.3cd "50 Gb/s Ethernet, 100 Gb/s Ethernet and 200 Gb/s Ethernet" Task Force, whose work is ongoing. The Task Force appears to be focused on creating a four-lane approach for 200-Gbps applications, which would imply four fiber pairs (one fiber in the pair for transmit, the other for receive). Meanwhile, the 400 Gigabit Ethernet PMD for such cabling was ratified last year along with the rest of the IEEE 802.3bs standard. That PMD calls for 16 fiber pairs based on 25-Gbps lanes – or 32 fibers per port. That's likely more fibers than some data center network architects would want to use.

Enter the "Next-Generation 200 Gb/s and 400 Gb/s PHYs over Fewer MMF Pairs Than in Existing Ethernet Projects and Standards" Study Group – or the NGMMF Study Group for short. The Study Group formed after a successful call for interest at an IEEE meeting in November 2017. Under the chairmanship of Robert Lingle Jr. of OFS, the group subsequently held its first formal meeting in January 2018. The group aims to create a Project Authorization Request (PAR) and Criteria for Standards Development (CSD) as well as recommend objectives that, if approved, a subsequent Task Force would use as its charter for standards development.

The initial meeting set the stage for this work via a variety of presentations that ranged from reports on the feasibility of various technological options to expressions of interest and guidance from large-scale data center operators who might benefit from more streamlined multimode fiber options for high-speed Ethernet networks.

Lingle, in a conversation with Lightwave, pointed to these presentations as indicative of multimode's bright future in data center networks. In one, titled "400G SR8 for Data Center Interconnect," Zuowei Shen of Google described his company's interest in using an eight-pair multimode-fiber approach as an alternative to the recently ratified 400GBASE-SR16 that's now part of 802.3bs. Shen envisions the use of transceivers (perhaps OSFP or QSFP-DD modules) operating at 50-Gbps PAM4 wavelengths, similar to the singlemode fiber PMDs from 400 Gigabit Ethernet. This approach would leverage available technology (with more PAM4-friendly VCSELs expected in the near future). For this reason, Shen theorized that SR8 could be deployed in volume this year.

A second presentation, co-authored by Lingle alongside David Piehler of Dell EMC and Chongjin Xie of Alibaba, explored the benefits of an approach they called "400GBASE-SR4.2" – a four-pair scheme that would leverage two wavelengths per fiber (hence "SR4.2"). In the presentation, the authors reported that Alibaba currently uses 100GBASE-SR4 multimode fiber connections in its network, which it finds more cost-effective than connectivity based on single-mode fiber and either PSM4 or CWDM4 optics. The cloud services giant would like to use a similar four-fiber approach to 400 Gigabit Ethernet in the 2019 timeframe.

Such a desire to reuse existing cabling is not unusual, based on Dell EMC's experiences. The company saw a similar phenomenon as network operators upgraded to 100 Gbps in the data center; in fact, Dell EMC could have sold duplex 100G multimode fiber transceivers as early as 2016 had such devices been available, the presenters stated.

And group member ambitions don't stop there. A presentation from Vipul Bhatt of Finisar urged consideration of duplex 400G as an objective.

Early progress

The January meeting included a number of straw polls and discussions of potential objectives for the two transmission rates. Lingle was reluctant to share details of the deliberations; however, the meeting notes and presentations posted in the Study Group's public area indicate support for both the SR8 and SR4.2 approaches to 400 Gigabit Ethernet. Consensus on an approach to 200 Gigabit Ethernet appears a bit farther off.

Along these lines, Lingle indicated the earliest the group would finish its deliberations would be at the next meeting, scheduled for early March. A Task Force could be formed as early as May 2018 if the Study Group wrapped up its work this quickly. However, the group's charter includes six months of deliberations, which Lingle noted could be extended by request if necessary.

Key technologies

In the 400 Gigabit Ethernet domain at least, approaches such as SR8 and SR4.2 imply the use of technologies not currently part of the 802.3bs multimode fiber specifications – 50-Gbps transmission and multiwavelength capabilities. However, both of these technologies are either already available or nearly so. Transmission at 50-Gbps via PAM4 is a key facet of 400 Gigabit Ethernet over single-mode fiber. Meanwhile, multiwavelength optical transceivers for multimode fiber, based on technology called shortwave WDM (SWDM), have already been unveiled (see, for example, "Finisar producing 100G SWDM optical transceiver for upgrades from 10 to 100 Gbps"). An industry group, the SWDM Alliance, has formed to promote use of the technology (see "SWDM Alliance targets data center wideband multimode fiber use"). A module multisource agreement (MSA) initiative has sprung up as well (see "SWDM MSA publishes two shortwave WDM standards"vtfytcweeqerfdsyrv).

Another technology that could apply to this area is OM5 fiber (originally known as wide bandwidth multimode fiber), which was designed with SWDM technology in mind. Early indications are that the Study Group members will focus on the use of OM3 and OM4 fiber in their deliberations. Still, OM5 fiber vendors clearly would be pleased to offer an alternative that likely would support greater reach than OM3 and OM4 would support.

Regardless of the technologies used, the creation of the Study Group and the presentations from the cloud services providers during the January meeting bode well for multimode fiber's use in data center networks, even in large-scale architectures at the fastest transmission rates available.

"Contrary to the hype about silicon photonics over single-mode fiber destroying the value proposition for multimode fiber in data centers, there is strong interest from both cloud and large enterprise data centers for short-reach, higher-speed links using VCSEL-based transceivers over MMF," Lingle concluded.

http://www.lightwaveonline.com/articles/2018/02/ieee-study-group-explores-future-multimode-fiber-roles.html