I believe this is the next big thing, to get rid of the bottlenecks where they really are:
Here are a few reasons for the boom in photonics.
By Steve Jurvetson
From the October 2000 issue
Why are carriers bending over backward to spend large sums of money on the unproven products of photonics startups? Look at what's happening. Demand for photonic equipment, which sends data on particles of light, is skyrocketing. Internet traffic on the backbone networks has been doubling every three months, and shows no sign of abating. Companies are laying fiber on any cross-country right-of-way they can find, from railways to gas pipelines.
With all this bustle, why are the optical backbones feeling the crunch? In a nutshell, we have a web of random access mapped onto a hierarchically deployed network. It's as if you were running a firehose through a funnel: the top of the network pyramid will choke worst.
Data networks have been deployed with a set of outdated assumptions about the patterns of traffic. Pre-Internet, computers were used primarily for computation and personal productivity. Today they are used primarily for communication. This shift has a profound impact on the ideal network topology. The vast majority of data access used to be from the local hard drive, with an occasional file or print service over a local area network (LAN). Occasionally, the departments would communicate with the enterprise server. Intercompany communication over private trunk lines was a rare activity. This was a classic "trickle-up" hierarchy with a strong set of statistical assumptions that the bulk of the traffic would be local.
Fast forward to today. The Web has blown apart the historical assumptions about traffic flows. The enterprise desktop communicates with computers all over the world. It's no longer a custom electronic data interchange project; almost every computer connects daily to another computer, with little sense of physical or network proximity. Web-hosted services and Microsoft Office 2000 continue to blur the distinction between local and remote files.
Application service providers, Microsoft.net, and, soon, streaming applications enabled by companies like AppStream will transfer more of the application logic from the PC hard drive to the network. It's a wild web of data access mapped to a neat presumption of hierarchy.
The LAN connection to the corporate desktop is not the bottleneck. In fact, the growth in photonic backbone traffic dramatically exceeds the sluggish upgrade of desktop connectivity. It's the pattern of traffic, not the capacity of the end nodes.
There are several bottlenecks up the networking chain, from wide area network to metropolitan connectivity, that are also by-products of increasing intercompany communications. Rather than purchase monolithic, mainframe-size telecom access equipment from the traditional vendors, modern carriers are buying highly adaptable and flexible systems from new companies like Cyras Systems.
These new companies are like the workstation vendors competing with the mainframes. Many of the shortcomings of the mainframe -- from glacial product development cycles to cumbersome application programming -- apply to the incumbent telecom equipment companies. As with all disruptive technologies, new companies are leading the charge.
Due to the basic physics of bandwidth capacity, photons are more effective than electrons for data transport. The photonics industry is is at the same level of development as the electronics industry was in the vacuum tube era. Companies like E-Tek Dynamics have mastered the black art of precisely hand-assembling discrete elements that need to be aligned with micrometer precision. The insertion loss budget, or optical throughput efficiency, of these systems is a concatenation of the losses that occur at each discrete interface.
Draper Fisher Jurvetson was the seed investor in Lightwave Microsystems, a company that is taking the next logical step: optical integrated circuits (OICs). This breakthrough offers the same sort of performance, miniaturization, and mass-manufacture cost benefits that earlier integrated circuits (ICs) offered the electronics world. Electronic and photonic ICs are built on silicon wafers with similar manufacturing equipment. Rather than aluminum wires, the OIC uses silica glass waveguides, and rather than doped polysilicon transistors, the OIC uses doped polymers as active switching elements. But both involve the mass-manufacture of multiple integrated devices on a silicon wafer. The discrete hand-assembled approach will go the way of the vacuum tube.
To date, most of the photonics industry relates to data transport. With OICs and micro-electromechanical systems (MEMS) mirror arrays, the all-optical switching function is on the horizon. Initially, active optical switches will be used as add-drop multiplexors for pulling optical off-ramps from the multilane highways -- the synchronous optical network (SONet) rings of the metro area. With optical switches, the functions of service provisioning and fail-over protection become much more efficient than the conversion from optical to electrical and then back to optical. A handful of these MEMS mirror companies were bought out for several billion dollars apiece, well before they had working prototypes.
These switching applications are for network deployment -- an infrequent activity that requires only millisecond switching speeds. The core switches of the Internet, the ones that route the packets of information at blazing speeds, are still basically electrical switches. The photonics industry is far from all-optical switches that match the speed and density of state-of-the-art electronics.
To understand the challenges in the core switching market, imagine that routing Internet packets is like trying to move taxicabs through New York City at rush hour. You could use one massive computer to run a global optimization routine and then radio commands to each taxi. This technique might work for small towns, but as you scale to larger and larger cities, you would need a networking supercomputer to keep up. This is the approach of Juniper Networks (Nasdaq: JNPR) and Cisco (Nasdaq: CSCO) and almost every current switch vendor. Without a paradigm shift, the core of the Internet will fail to scale.
Enter the entrepreneurs again. Brightlink Networks is pioneering a different and infinitely scalable approach. It lets each cab make local optimization decisions based on local congestion (e.g., Fifth Avenue looks badly backed-up; I'll try jogging over to Third). The decisions are local, the algorithms can run very quickly, and it doesn't matter how big the city is. It's a system topology that maps well to the Internet. Equipment is not discarded during upgrades; rather, more taxis, or switch chips, are appended as the system grows. (Of course, it's a bit more complex, since BrightLink isn't dealing with two-dimensional surface road maps, but rather a multidimensional hypertorus, lots of intersecting rings of connection.)
The consulting firm RHK estimates that the market for optical core transport systems, which didn't exist in 1999, will reach $7.3 billion by 2003. Carriers need to spend, and spend big, on these systems to avoid bottlenecks at the top of the networking hierarchy and keep the Internet running at rush hour.
Steve Jurvetson is a managing director of Draper Fisher Jurvetson. |
