100G Optical Modules for Data Centers? Let us count the ways

Number of hits：1892016-09-20 00:00:00　source: SHENZHEN TONGSION TECHNOLOGY CO., LTD

Where we’ve been

In 2010, the IEEE 802.3ba project completed the first three standards for 100G Ethernet: 100G - SR10, 100G - LR4 and 100G - ER4, with reaches of 100m, 10km and 40km respectively. The 802.3bm Task Force added 100G - SR4, but in 2013 was unable to reach consensus on a 500m PMD, a sweet spot for large data centers. Industry moved quickly to fill the void with multiple MSAs, with 100G - PSM4, 100G - CWDM4 and 100G - CLR4 all finding success in the market. A number of special products for have also been developed for individual customers, notably ‘LR4 lite’.

What’s happening now?

Modules based on CWDM technology are seen as lower cost than modules using a LAN-WDM grid, so the CWDM4 MSA group is moving to add 10km, 20km, and 40km variants to the current 2km MSA. And the ITU-T Study Group 15 has defined a 100G - ER4 - Lite (standard 4L1 - 9D1F), a more practical variant that assumes an APD receiver instead of using an SOA. If you’re counting, we’ve now listed 12 variants of 100GbE modules.

2016 is the year for volume ramp of 100GbE modules for datacenter use. It’s also the ramp of the 25Gb/s serdes ecosystem, with network switches using ASICs such as the Broadcom Tomahawk. Is it time to start defining a new generation of 100GbE modules based on a 50Gb/s serdes? Demonstrations of “100G-CWDM2” modules at OFC suggest one possible new direction.

IEEE standards at multiple crossroads

In previous research notes, we described the trend in the IEEE to define optical PMDs for single-lane and multi-lane as per-lane speeds move from 10G to 25G and 50G. The 1-lane/4-lane model worked for 10G/40G and 25G/100G, so 50G/200G seemed the obvious next step. When you combine the possible N-lanes by Z-Gb/s with all possible reaches, however, you get a very large matrix of solutions. The new 802.3cd Task Force was chartered to define standards based on advances in 50 Gb/s signaling technologies and the original focus was for server connections. Hence the initial objectives were for 50G backplane for blade servers, twinax and MMF PMDs for rack servers. 

In May, two 2-lane (2 x 50G) objectives were included to support “next-gen” 100GbE on SMF. One 100G SMF objective, however, was curiously non-specific with regard to the number of lanes. It simply read “SMF with lengths up to at least 500m”.

To the surprise of some members of the task force, we once again have a debate on making the jump to 100G serial on a single wavelength.