Over 30 years ago, there began discussions in the industry about the development of an all-optical switch. Some of the comments sound familiar to those of today. In April, 1983, The Economist addressed them: “[Optical] transmission performance is degraded by inefficient switching via electronic devices at either end of a system.” Also, there was the statement that one school of thought favors an “in incremental or hybrid approach.” Although there is not much hype in the market right now about the widespread use of such devices, both major vendors, Huber+Suhner Polatis and Calient Networks, long-time survivors since the big bubble burst at the turn of the century, are expressing a great deal of optimism, particularly relating to at least three of the big four hyperscale data center operators, Amazon, Facebook, Google, and Microsoft, supposedly taking a serious look at their systems. However, our very latest intelligence gathering suggests that these Web 2.0 behemoths may be more attracted to a different type of technology solution.
O-E-O cross-connects were preferred over O-O-O systems starting in the early part of the last decade because the latter was too expensive and service providers required grooming down to the electrical STS-1 layer. There were other all-optical switch companies around back then, like Corvis, which do not exist now. At present, we believe that Glimmerglass Networks is struggling mightily just to survive.
Usually a lot more positive rhetoric precedes the entry of a new solution in a big way in the high-end data center space. With the exception of Amazon, the other members of the big four are not exactly bashful about promoting new technology solutions, which they may or may not buy, because they want as many options as possible. There are also at least a few network engineers in the industry that are certainly in a position to make an objective assessment, who will more than suggest that there is a lack of familiarity with all-optical switches by these companies. There has also been the characterization of these switches as being too slow for at least certain applications.
One of the problems with the two major vendors making further penetration is that they are in many cases competing head to head with cheap and dirty patch panels going for $30 or $40 a port, when these switch suppliers may be still selling at as much as an order of magnitude higher per port. If there were greater competition from optical switches, one can reasonably assume the prices on these panels would not remain static either – and it is fairly clear that both of the prominent vendors cannot be profitable in competing even with the current pricing levels of these panels.
In actuality, there are signs that the big hyperscale players are closely examining something vastly different, the concept being offered by Rockley Photonics, which is a wavelength router with a switchless architecture, as explained here. Computing the path is accomplished by choosing the wavelength.
Although Rockley is well-funded, we understand that certain proprietary and single-sourced components could be a major impediment in customers buying its product.
[written by Mark Lutkowitz]