Despite the unequivocal remarks on not currently planning to use 400ZR+ or necessarily for that matter, even higher capacity, inter-data center ZR modules by Ashish Vengsarkar, Head of Optical Networking Technologies, Google, USA, as the keynote speaker at the Data Center Summit of OFC 2022, there strikingly appears to be very little impact on the public discourse in the industry on such pluggables. Once again, Google represents the biggest force when it comes to the optics technology direction for the hyperscalers, and one would think that at least adjustments in the questions being asked in the space would be affected suitably. While Vengsarkar kept the door open to the idea of achieving the necessary link budgets including for 400-gig ZR+, 800ZR and “1.6T-ZR,” at least for the first type of module in this list, fibeReality’s intelligence gathering indicates there is no interest in it at all at the gigantic Internet search provider. Obviously, a more dramatic impact on potential cannibalization with established optical network equipment would occur with the longer reaches provided by 400ZR+, and we understand that Google is principally striving to avoid the management of too many systems. We also realize that the webscaler considers it to be a low-volume application and that the cost to create the OSS to support such one-off use cases is not worth the small savings. Actually, the major deployment of 400ZR+ gear is expected to occur at telcos, and therefore, the anticipated volumes would be substantially lower as well as much slower to develop, given the nature of more traditional types of operators. Regardless, the prevailing assumption in the hyperbole for this new product does not reflect that reality, most notably an overemphasis on usage in data center interconnect. All in all, this situation will continue to demonstrate that once a large investment is made in a particular narrative by the optics industry, especially encompassing high stakes in other related areas, any alterations in the established messaging will be resisted.
Concerning the webscalers and 400ZR+, there is the special case of Amazon, which has a sizable number of distributed data center networks, and its campuses are spread out to a great extent. The large cloud provider is the leader of the hyperscalers from an operations/cost point of view, as we noted at an EPIC event.
However, if a vendor can meet Amazon’s requirements, fibeReality perceives that in this case, interoperability is not considered a deal breaker. Perhaps if the company chooses to use bookended ZR+ solutions, it would achieve better performance anyway.
Conversely, despite Microsoft having an even greater attraction to 400ZR than has been the case with Google, the former may even match the latter over the dislike of 400ZR+. Microsoft has utilized ZR in a similar way to Marvell/Inphi’s ColorZ — most specifically pointing to multi-vendor sourcing purposes, particularly for its single-span links. Despite the distance limits of ZR, and regardless of past rhetoric from Microsoft promoting 400ZR+, we expect it to employ standard DCI.
Similar to Google, Microsoft is pushing for a 1.6T version of ZR with a reach of about 100 kms or less with requirements for the same volume, etc as 400ZR. Nevertheless, if a particular scenario develops, then the realization of a “1.6T-ZR,” in the space may be questionable, and somewhat paradoxically, again given market hyperbole (in this situation about high data rates), the necessity for low-end ZR devices could take on even greater importance. Yet, regardless of the apparent resistance against 400ZR+ for any application at both Google and Microsoft, it should be noted that other cloud provider types of firms might not rule out ZR+ for longer-haul connections as contrasted with their DCI apps.
For the more conventional kinds of service providers, in addition to making adjustments for longer reaches, there will be a wider range of requirements that fall outside of the performance window of the ZR. Most of them are waiting for the new Bright versions of ZR+ with high TX power and 200G 16QAM in 50GHz spacing, providing the best use case in brownfield deployments as well. At the same time, the added complexity in the entire process might make the cost difference with the standard version reach a significant amount (surely less on big volumes).
Returning to the major point of this piece, when telcos are ostensibly demonstrating a higher level of flexibility about employing 400ZR+ than firms like Google, there should be the expectation of red flags being raised. In fact, it has often been standard practice in the trade by everybody up to this point to use the shorthand of ZR/ZR+ in that combined way, despite the customer dynamics for each being so diverse. Of course, specific market research firms will undoubtedly persist in combining the two components together in reports because the projections of the total opportunity can be made to look larger, while disingenuously obscuring the principal demarcation point.
Suppliers of 400ZR+ gear will also likely keep on walking a fine line. Although testimonials will be collected stressing the “telco” usage of the equipment as an effective advertising strategy, they will strive to avoid categorizing it under that major rubric, such as on investor calls, so as not to voluntarily place limitations on the projected ROI.
Putting aside the potential for a global economic recession, there is a lot riding on the ZR+ narrative because it feeds directly into other bigger and highly exaggerated/entrenched marketing/PR types of stories, such as 1) a relatively quick transition to IPoDWDM (including at its extreme, the coming elimination of a separate optical transport network supported by OTN), 2) widespread openness of public networks, as well as 3) aggressive disaggregation (along with moving speedily away from transponder-based gear). To readily admit that the success of 400ZR+ will be heavily dependent on a bunch of what we affectionally call “ex-Bell-heads,” (which are at least to some extent, still dealing with their IP vs optical transport fiefdoms) in which the promotion of SDN still is predisposed to be the last example in what we have called a parade of panaceas going back to 1975, will be extensively thwarted for as long as possible.
In closing, the overwhelming propensity involving the term, 400ZR+, will continue to be spoken of linking data centers without the proper context. In the meantime, a main preoccupation of the large hyperscalers, driving by far the greatest amount of bandwidth growth in the world, will be in obtaining coherent solutions in the future, addressing shorter distances (10 kms) to the DC.
To comment on this article, please click here.