The highlight of the Ethernet Technology Summit held in Santa Clara, CA last week (April 30-May 1) was the presentation by Brad Booth of Microsoft titled “Building a Castle on a Cloud.” Brad shared his vision of future datacenters, where the equipment, including optical transceivers, is replaced every three years. One would ask, why should datacenters not get upgraded as often as consumers purchase new iPhones? According to Brad, Microsoft is getting close to purchasing 100,000 new servers per month now and will need a massive amount of 100GigE connectivity by 2015. Is there a solid ground for this vision or is it just “castles in the clouds”?
The final version of our Quarterly Sales Database, delivered to LightCounting clients this week, offers a detailed snapshot of the market today. Sales of Ethernet optical transceivers were up by almost 30% in 2013, boosted by demand for 10GigE and 40GigE connectivity in datacenters. Demand for these products has remained strong in early 2014.The big question is - will massive deployments of 100GigE be one year away or are they more likely to be five years away?
The variety of technical approaches and MSAs for 100GigE datacenter connectivity is mindboggling. Despite this confusion, transceiver vendors should be able to offer viable 100GigE products, such as QSFP28 modules, by 2015. Will Microsoft purchase more than a few thousand of these modules by 2015?
More than half a million 40GigE modules were sold in 2013, and potentially all these connections will use 100GigE, as well. The main problem with this transition is that all 40GigE modules used in datacenters combine four lanes of 10GigE into one module, fitting seamlessly into the vast base of 10GigE connectivity while improving port density. Massive deployments of 100GigE (4x25G) modules would require a large base of 25G switching silicon in the datacenter, which does not yet exist.
Microsoft’s vision is consistent with this argument and plans to rely heavily on 25GigE connectivity, ignoring IEEE standards. If Microsoft’s plans materialize, Facebook, Google and others will have to respond. Competition among the Internet giants is brutal and technological superiority is part of their business strategy. The right question is whether Microsoft customers will see the benefits from having 100GigE in their datacenters and how soon it will happen.
A related question is whether Google Fiber customers see benefits from 1Gbps broadband access and are ready to pay for it. According to Netflix, their Ultra HD 4K videos are delivered smoothly over 20 Mbps lines, so 1Gbps would seem to be overkill. Yet, AT&T is just as aggressive as Google now in marketing 1Gbps connectivity in selected US metropolitan areas. It appears this may be a case where the fear of losing business to competitors overshadows solid technological planning.
The problem with markets and technologies driven mainly by competition is that most of the demand is based on the expectation of future business, which can change abruptly. Forecasting these markets is not an easy task, but that is exactly what the LightCounting team is working on now.