At OFC 2023, the real buzz should have been about the mounting public support by vendors of the form factor, OSFP-XD, which will ultimately lead to a substantial amount of revenue for the marketplace in the future. Additionally, what could easily be perceived as an apparent repudiation of co-packaged optics in favor of pluggables by Marvell Technology (in the past, a prominent supporter of the former concept) on its quarterly financial call, just before the large conference, seemed to be totally ignored at the show. Rather, industry cheerleaders chose to get behind another unfounded narrative, this time involving the idea of Linear Drive Optics Modules (LDOM) that could potentially force another wasteful expenditure of resources, most significantly by the well-established, but heavily margin-conscious, pluggable vendors. Despite rhetoric to the contrary, similar to the misguided, substantial push for CPO, and earlier, COBO, the notion that the elimination of DSPs is a novel idea, or lacking a basis in historical precedent, would be incorrect. The LDOM story also is an offshoot of another misleading portrayal in the space that substantial power reductions in the data center can be achieved by making modifications in the optics gear (especially as compared with other devices). Most significantly, LDOM is conspicuously about the use of marketing ploys on the part of a couple of prominent players, but with opposite goals in mind. Broadcom has hardly anything to lose in promoting the concept. Yet, for Arista Networks, fibeReality believes there can be only one explanation for what otherwise appears to be an illogical advocacy of LDOM — to outfox the promotional efforts made by CPO proponents with direct drive (including Cisco Systems — although there is hardly universal agreement even there because many individuals at the vendor understand that they will be grappling with the lack of reliable specifications).
Ironically, it is once again, at the OFC conference itself, in which inadequate attention is given to responses by important participants, as happened with the matter of 400ZR and 400ZR+ a year ago. During the 2023 Market Watch panel session on “PAM vs. Coherent for Data Center Connectivity,” the organizer, Lian Qin, Marketing AVP at Marvell asked the speakers about the importance of LDOM. After Cedric Lam, Principal Engineer at Google, stated that although “the intent is good,” he was “concerned about reliability,” Lee Xu, EVP at Coherent responded that he was also seeing issues with linear optics. In short, it was fairly clear to the audience that neither of them thought it was a viable option.
Subsequent intelligence gathering from Google by fibeReality further bolstered the position of the influential, large user of optical gear. Based on its internal investigations along with its constant involvement in achieving robust link designs, the use of DSPs is simply inescapable at high data rates. Avoidance of a good number of interoperability matters that comes with linear interfaces was stressed to us as well. Moreover, we see it as another potential example for Google to fill the vacuum that developed in the hyperscaler realm to more openly press for, or to urge against, various optics solutions under consideration.
At Coherent’s Analyst Briefing, CTO, Julie Sheridan Eng, was much less emphatic about the negative attributes of LDOM. However, she wisely mentioned the past technical difficulties with coherent ACO and how much the DSP in the module became an enabler. (An even older lesson occurred with 10G-LRM that engineers to this day are still suffering from shell shock, and so, the prospect of a couple of decades later taking on a linear interface with PAM4, with its very complex coding arrangement, is certainly unattractive, to say the least.)
Turning the attention to Broadcom, one should take note that the firm still has difficulty making the case for CPO, albeit, it never seemed to be serous about the prospects at the highest executive levels anyway. In contrast, with LDOM, the vendor seemingly has everything to gain in taking further advantage of its dominant position with switch ASICs. Although in working with Broadcom, MACOM did well for itself in putting together a few demos at OFC, it is a far cry from analyzing actual link design needs. Furthermore, Broadcom may want to be a little careful when it comes to its zeal in advocating for the eradication of particular components, especially when it involves lower power utilization, as it is not out of the question that the player might one day face the prospect of certain measures being adopted that could come very close to home.
Upon a superficial look at the advocacy of LDOM by Andy Bechtolsheim, Arista’s Chief Development Officer and Founder, while he would remain in the position of the number-one, outside spokesperson on OSFP-XD for Google, he is evidently advocating a stance on this latest scheme that disagrees with the big equipment purchaser. However, upon reflection, a reasonable conclusion is that Google is probably ambivalent, at worst, about Andy’s stance because of the disproportionate, unwarranted amount of attention given to CPO’s direct drive. Although he has advocated for ways to reduce power consumption in the past, the idea that somebody with such intimacy on network requirements would recklessly discard the DSP, and then put designers in an untenable position of cleaning up a tremendous mess, seems nonsensical. Even at the Optica Executive Forum and at an earlier OFC workshop, Andy managed to convey a clear understanding of the unacceptable hurdles involved with LDOM in the following two ways: 1) stressing the requirement for “very careful signal integrity design” (then paradoxically adding to what is supposed to be a promotional effort, “any impairment to end-to-end channel affects BER performance;” and 2) mentioning the need for “a high-performance switch SERDES” (arguably suggesting the very real barrier in scaling to 200G PAM4). For the most realistic picture of LDOM from Arista, one should look at its post-deadline paper at the conference.
In closing, on our previous references to Nubis Communications, our primary concern was on the improvements in its marketing message, which fibeReality considered to be originally scattered. Since that time, we obviously had the opportunity to substantially study the specific merits of its focus on “direct drive.” Although our preliminary assessment was that it could be attractive for active optical cables, where interop would not be a problem, we later learned that even for this application, there will be difficulties to overcome at the system host interface regarding reflection, crosstalk, and other relevant matters.
To comment on this article, please click here.