Industry Voices—Gasman: QSFP at the end of the road?

A data center
Gasman CIR

I have been analyzing market opportunities in data communications for more than 30 years, and have concluded that end users will always opt for tried-and-tested technology until it simply isn’t viable any more. Call it the principle of data center technology conservatism, if you will. Data center managers will show considerable interest in that next big thing. But when it comes to spending the dollars, they will go for the sure thing. This is understandable; for data center managers, an entirely new technology direction is a career-destroying move if it doesn’t work out.

Under the principle of data center technology conservatism, technologies are often pushed to the edge of their capabilities. Copper-based InfiniBand had to literally fall through the floors of data centers before data center managers would give up on copper. For 400G data center connectivity, end users are now being presented with new embedded optic technologies, primarily in the form of the Consortium for On-Board Optics (COBO), which promises an MSA-like approach to embedded optics which will kick in at 400 Gbps. How well can we expect embedded optics to do in the market?

Based on the technology conservatism principle outlined above, CIR’s recent report, “The Embedded Optics Market: COBO and its Alternatives—2017 to 2026,” suggested that the popularity of the QSFP MSA will keep QSFP (perhaps in the form of QSFP-DD) viable well into the 400 Gbps era. 

The long history of QSFP in the data center coupled with its broad support of media, optics, cabling, reaches, relatively low costs, etc., are powerful arguments for QSFP’s long-term survival. Today, QSFP is used in just about every data center and in its QSFP-DD form factor can enable up to 14.4 Tbps aggregate bandwidth in a single switch slot. And QSFP is not going to fall through the floor, as it were, any time soon! 

One might conclude from all of this—via the principle of data center technology conservatism—that COBO is ahead of its time. CIR’s analysis suggests that COBO won’t generate large revenues until 800 Gbps and 1 Tbps intermachine connectivity begins to be significantly deployed. On the other hand, unlike most MSAs, COBO is a creature of a large end user, Microsoft. Microsoft can therefore presumably be counted on to buy into COBO for its huge cloud data centers once COBO boards are commercially available, which will be in 2018 some time. In the past, end users haven’t had much input to MSAs.

Other large end users may follow Microsoft’s lead in embedded optics as early as 2018, pushing up the prospects for COBO/embedded optics. (The prime disadvantage of embedded optics—that boards come fully populated—isn’t a problem for the largest cloud providers). However, QSFP isn’t really going to lose out to embedded optics/COBO until it metaphorically falls through the floor. This will happen as the data center is stress-tested by the rapidly growing needs of bandwidth-hungry applications. Data centers are already full of 4K video, but this will soon start to migrate to 8K video. Traffic from the incipient internet of things is turning from a trickle into a torrent. 

These rapidly emerging trends will place pressures on both the QSFP environment and embedded optics/COBO as early as 2019. And we think that in the final analysis, the latter will win. The reason is that embedded optics is inherently better able to cope with the increasingly demanding requirements for cloud center data communications. 

The two requirements that really stand out in this regard are power consumption and transmitter density. Here we note the inherent superiority of embedded optics over pluggable optics. In embedded optics, as opposed to pluggables like QPSX transceivers, the optics and electronics are not confined within a mechanical enclosure, and therefore power dissipation is less of a design issue. Transmitter density—and hence costs per bit—can also be boosted in embedded optics to a greater degree than in QPSX designs as the result of less packaging and higher levels of integration (eventually to the point of co-packaging design strategies).

The point here is that when QPSX finally does fall through the floor, embedded optics, COBO-derived or otherwise, will be there to take data center optics in a new direction.

Lawrence Gasman is president of CIR. He has covered both the optical networking equipment and optical components market, and his recent market analysis work has focused on optical data centers, embedded optics, 5G infrastructure and quantum networking.