Our Optical Realizations in Bordeaux

Avatar photoPublished on
shutterstock 1688555299 copy

It was so much fun to speak in person at ECOC 2021 and directly make eye contract with individuals in a live audience rather than doing it through the impersonal nature of a Zoom conference. This writer immediately stated the inherent uneasiness in comparing two solutions head to head, OSFP-XD and co-packaged optics, which both will not be significantly real for over five years. Trade shows should spend more time on minor modifications of mature products, which are more interesting to discuss than gear, which may never see the light of day. Nevertheless, not all vaporware are created equal. In being part of a long-term, evolutionary track of other pluggables, one can have a reasonable amount of confidence that XD will overcome its technical obstacles to become the seventh, major, high-volume form factor; whereas CPO is so revolutionary in nature, it would require an uncommon aberration to occur (which by definition cannot be predicted) to turn into a significant part of the intra-data center network. This article provides fibeReality’s latest insights and further elaboration on past points made in “OSFP-XD Takes the Stand,” in June, and which were stressed in our talk at the ECOC Market Focus session. In future pieces on our blogs, we will also discuss two new matters that came to our attention during our informal discussions on the exhibit floor (as such activities lead to a wealth of intelligence) including: 1) a potential shakeup in the Microsoft optics networking management and 2) Avicena’s novel (and rather impressive) means of addressing up to 10-meter applications in the DC as well as other markets (while trying to determine the drawbacks of such an approach).

In Bordeaux, our latest viewpoints in comparing OSFP-XD with CPO included the following 11 items:

1.After signal integrity, which is consistently a worry with any kind of high-rate transmission, cooling is the second most vital issue with XD as the heat load will be greater than an OSFP. Neither a relaxation by hyperscalers in temperature standards nor unusual thermal technologies are expected. fibeReality’s latest intelligence points to the only need being a more than 1RU card spacing (the height may have to be 1.5- or 2RU to support cooling).

2. The hurdles with connectors are hardly impossible to tackle and there appears to be a great deal of confidence at companies like CommScope, based on our intelligence gathering, in the arrival of an XD.

3. There is definitely no precedent by webscalers to spend more money on hardware to save electricity cost. There seems to be two separate groups involved with the management and maintenance of each budget with no communication between them (the tax implications may be distinctive as well). The actual gear expense is the dominant criterion, not the power cost.

4. From an engineering standpoint, the XD offers a solution that cannot be accomplished by a QSFP-DD for a 102.4T switch. We are now leaning just based on logic beyond a 50-50 chance that a pluggable will be used for a 204.8T solution because of the historic trend of finding ways of avoiding the replacement of progressively shorter electrical connections with optical cable links as rates are changed.

5. Facebook, a big supporter of CPO, as with Microsoft, leaves something to be desired on homegrown notions in optical such as its irrational push for 400GbE-FR4 in QSFP-DD. The former also never seems to be against any new idea of any kind, such as its support for TIP for incumbent operators. Moreover, Facebook’s business model requires relatively low-capacity amounts in its network as it is only a social media company, while not providing cloud or retailing services — making it in the least amount of in a hurry of the big four US hyperscalers to require XD or CPO. Furthermore, its resistance to starting off with a simpler play, at least in the past, with the use of multimode co-packaging is hard to figure out.

6. The most favorable statement that can be made about CPO is that it is still running late in its course. Despite all of the present job offerings for design talent on LinkedIn, some optical engineers still perceive the concept as the last standardization foray until their retirement and they have a vested interest in keeping it going regardless of its future utility. A fallout is inevitable as the facade becomes even clearer — and it is also the fear for these people in these standards activities that CPO will not be all that it is cracked up to be and will fail to be fully realized.

7. In the meantime, there can be over 100 people on an OIF teleconference regarding CPO and everyone is just tending to watch the others with little to no work being accomplished. There are credibility issues just with the idea of NPO being up for grabs on definition. Is it actually CPO or more specifically a comeback of SNAP12 — or even rebranded COBO? A lot of what the OIF is doing in CPO will eventually die.

8. fibeReality does not take any supposed development work by Broadcom on soldered attachment of optical engines seriously other than perhaps for purposes of providing demonstrations. We are not expecting the building of such gear. In fact, we would now not be surprised if the supplier gives up its silicon photonics group around the arrival of 51.2T switches given the lack of adequate margin generation.

9. The “we’re going to do everything with optics” mentality goes back decades and the prevailing outcome will be no different with CPO. Optical storage was supposed to replace the magnetic variety a very long time ago. Optical computers were projected to happen by the year 2000. All-optical switches were thought to take over the public network world. Despite Rockley Photonics’ IPO, the search for the killer app for SiPh has continued. With 5G, the hype about microwave over fiber has re-emerged. Optical FPGAs have never gone beyond proof of concept.

10. Existing technologies, such as pluggables, usually find ways to extend their lives. One just has to look at the survival of asynchronous M13 multiplexers so many years after the synchronous revolution. Additionally, the TDM-IP shift is still happening.

11. Intelligence gathered from Microsoft indicates that there is interest in the XD, but more data is needed regarding applicability at higher rates.

Other issues that we will also address in future blog pieces pertaining to our actual ECOC presentation include the increasing number of voices at the webscalers, a certain amount of preliminary evidence of a rebalance starting to occur between optical suppliers and customers, the hyperscalers and the supply chain, how the time element helps XD compared to CPO, the incompatibility of Microsoft with hardware vendors, and comparisons with the large Chinese DC operators.


1 comment

Comments are closed.