The Path to 100G Single Lambda in the Data Center - Part 1
It is pretty widely acknowledged that 2016/2017 is prime time for 100 Gbps short reach optical links in the data center, interconnecting servers and switches. While this is true, more accurately, this moment in history is prime time for 25 Gbps per wavelength or "25G single lambda" technology commercialization.
For example, the QSFP28 MSA transceiver form factor implements 100GBASE-SR4 short reach (up to 100 meters) links using 4 x 25 Gbps parallel optics, typically on OM4 multimode fiber. It also implements 100GBASE-LR4 long reach links using four channel course WDM-based (LAN-WDM) serial optics, on single-mode fiber. These are both direct detection systems, using simple on-off keying / intensity modulation. The entire 25G single lambda signal chain, from the electrical SERDES I/O, the clock/data recovery retimers, the limiting drivers and laser diodes, the photo diodes, and the transimpedance amplifiers, i.e., all the necessary ingredients for 25G single lambda optical transceivers have come together to enable successful 100G commercial deployment in data centers and other networking applications.
This blog post is the first in a series on how the industry is likely to transition to 50G single lambda pluggable short reach transceivers (making 200 Gbps optical Ethernet transport commercially viable) and eventually to 100G single lambda pluggable short reach transceivers (making 400 Gbps optical Ethernet transport commercially viable). Indirectly, this blog series investigates whether coherent detection methods will find a role to play in short reach optical links as rates approach 1 Tbps, as well as the future of multimode versus single mode fiber in the enterprise data center.
One of the "boundary conditions" that underpins this analysis (and one that has been widely acknowledged by industry analysts) is that, all things equal, network operators prefer to reuse existing optical cabling infrastructure when making data center network performance upgrades. This preference obviously does not apply for new data center build-outs, but as a large part of the market serves existing data center upgrades, this thought process tends to float to the top of mind of data center designers and operators. There are a couple of important ramifications of this preference.
First, new short reach transceiver technology must maintain the minimum reach of the previous generation(s), typically 100 or 125 meters.
Second, the number of optical lanes must be consistent with previous generations. Specifically, the industry has shown a preference for four (simplex) / eight (duplex) fibers and MPO connectors since the transition from QSFP+ (4 x 10 Gbps) to QSFP28 (4 x 25 Gbps). There is an economic sweet spot for eight fibers per link in the data center, and a strongly established ecosystem, including 12 ferrule MPO/MTP optical connectors (along with the requisite associated cleaning and inspection tools). This sweet spot makes the success of 32 fiber proposals such as IEEE 802.3 400GBASE-SR16 (16 fibers x 25 Gbps x 2 for full duplex) questionable.
Third, the new transceiver technology must not require new fiber types, such as WBMMF (now OM5), unless building out a new data center. Specifically, for short reach data center links, the new transceivers must provide adequate performance on legacy OM3/OM4 multimode fiber.
We'll explore additional interesting angles of these technical and commercial dynamics next time.