PDF of this tutorialOptical Ethernet Switches
A true “all-optical network” is not likely to happen in the Ethernet space in the foreseeable future. The reason for this is that there are no technologies that enable packet switching at the optical layer today. However, purely optical Ethernet switches have already been demonstrated, in the sense that all of the Ethernet interface ports are optical (while the internal switching remains electrical). This is today’s technology optimized for carriers selling Ethernet services. When a carrier’s customers are spread over metropolitan distances (10 km – 50 km), then the lowest-cost service-provider network is Ethernet with all of the device ports themselves being optical. Also, businesses are requesting native Ethernet services to interconnect their facilities into VLANs. This trend will be accelerated by the fiber-to-the-consumer move, as the lowest-cost optical service will once again be Ethernet. So the bulk of early fiber-to-the-home (FTTH), and possibly the majority of fiber-to-the-business (FTTB) services, may well be optical Ethernet (at least as viewed by the carrier).
GBIC Modules
The latest GBIC innovation is a mini–GBIC, also known as the small form-factor pluggable multisource agreement (SFP MSA) module. The mini–GBIC is only about half the size, effectively doubling the available capacity that can be designed into the face of an equipment shelf. New designs are quickly taking advantage of this space savings.
Resilient Packet Rings (RPR)
While not formally a part of Ethernet, the IEEE 802.17 committee is creating a standard for packet transport over fiber-optic rings. Current directions include using Ethernet framing in SONET–style rings. The goal of RPR is simple: to define a high-performance, high-availability optical transport suitable for carrier networks in metropolitan service areas. Note that it is easier to implement a fast and robust link-failure recovery in a ring topology than in a mesh topology; this is because in a ring, the alternate route is always known.
The IEEE 802.17 committee does not view RPR as Ethernet, and indeed there is no intent that an IEEE 802.3 device could be directly connected to an 802.17 interface. To a degree, a metro Ethernet is a competing mesh (not ring) technology with relatively slow spanning-tree failover, instead of SONET–style, fast (50 ms) failure recovery. Still, RPR implementations are likely to become a popular transport mechanism for Ethernet packets, especially for telecommunications service providers.
10-Gigabit Ethernet Proposed Standards
The 10GEA (10-Gigabit Ethernet Alliance) is an industry consortium of about 100 members working to promote the acceptance and success of 10-gigabit Ethernet. This group is not the same as the IEEE 802.3ae standards committee, which is working on a set of proposed standards for 10-Gigabit Ethernet. The manufacturers are not passively waiting on a standard, as multiple vendors expect to announce 10-Gigabit Ethernet products during 2001, in advance of the standards ratification expected in March 2002. Of course, each of these vendors hope that their products will be one of the standards, and they promise that they will conform once those standard interfaces are defined.
10-Gigabit Ethernet May Be Optical Only
There are real challenges to making electrical signals carry a significant distance at those rates, and the implementations may simply be more expensive than their more capable optical cousins (meaning that they simply won’t happen).
One of the proposed standards uses very-short-reach optics, to be implemented as parallel data streams over a fiber-optic ribbon containing 12 multimode fibers. This was proposed as a low-cost method to interconnect devices in a room.
A second proposed standard uses a very compact package (about 1" x 0.75" x 3") containing a coarse WDM device, four receivers, and four lasers operating approximately 25 nm apart in wavelengths near 1300 nm. Each transmitter/receiver pair operates at 3.125 gigabaud (data stream at 2.5 Gbps). The proposed device has very aggressive engineering challenges and a very aggressive target-price point.
A third proposed standard is a serial interface using 64B/66B encoding (instead of the 8B/10B used in gigabit Ethernet), a data stream of 10.000 Gbps, and a resulting clock rate of 10.3 Gbps. This is a favorite of the Ethernet “purists,” who like the simplicity of just moving the decimal place; moreover, the resulting optical data stream may be directly interoperable with some of today’s DWDM systems.
A fourth proposed standard is a SONET OC–192 compatible stream, which is therefore clocked at 9.953 Gbps. The disadvantage is that it is not a pure 10 times 100 megabit Ethernet. There may also be disadvantages of cost—SONET is not the least expensive transport method. The advantages, however, are that it would be guaranteed to interoperate with all of the OC–192 SONET devices, including the networks of all of the major telephone companies, and all of the OC–192 compatible DWDM systems. And the reality is that 9.953 Gbps is close enough to 10.000 that no “real” applications are likely to notice the difference.
Figure 4. Bandwidth-Growth Timeline
