PDF of this tutorialOptical Ethernet systems are evolving beyond mere “optical links” that interconnect isolated LANs. Rather, they are becoming systems in themselves, providing scale and functionality that is simply not feasible with copper-based Ethernet, including those linked by routers.
LAN
Today, few optical Ethernet links are implemented within a computer room or small building. But there are exceptions for electrically noisy environments, highly secure transmissions (no EMI), and ground isolation. And even in a small building, it is easier to run fiber-optic conduit than electrical wires because there are fewer issues with building codes.
This situation is likely to change, as very-short-reach optics support much higher speeds than copper does. It is important to note that gigabit copper links are limited to about 30 meters and that the next generation of Ethernet at 10 Gigabits would drop this already inadequate distance dramatically.
CAN
Today, virtually all Ethernet links greater than 200 meters are implemented optically. The CAN is dominated by multimode fiber, although most CANs are really multiple LANs interconnected by routers that use optical links. This situation is changing, as the scale and functional capabilities of Ethernet switches increase. Ethernet 10BASE–T hubs, once dominant, no longer offer a sufficiently lower cost to justify their use. More and more LANs are being implemented with Ethernet switches, providing separate switch ports to every node on the LAN. Traditionally, these Ethernet switches aggregate traffic into a high-speed uplink port (once Ethernet, then Fast Ethernet, and now Gigabit Ethernet), which feeds a router that itself interconnects the LANs and provides WAN access. But today, when the cost of a router is weighed against the cost of a GBIC module, it loses every time. This trend is being accelerated by the proliferation of virtual LAN (VLAN)–capable Ethernet switches and by the development of even larger and more capable routers that understand and route to VLAN IDs in the now larger LAN.
MAN
Optical Ethernet in the MAN is a relatively recent development. Gigabit optical Ethernet has the capacity to provide direct Ethernet services as a carrier offering, with service switches that limit actual delivered bandwidth as needed. Multiple vendors now offer direct Ethernet services to subscribers, with only a few core routers linking those subscribers to the outside world (the Internet).
But “Ethernet services” is not the leading reason to implement an Ethernet MAN today. Rather, the desire to reduce the number of routers in the network is becoming the most compelling reason to use a Layer-2 technology (Ethernet) in the metro area. The reason is simple—every router in a path, many of which are unnecessary, adds delay to the packet transport. In an ideal world, routers provide a buffer between management domains, which are generally companies or network providers. They provide a place to control access, provide security (via firewalls) and manage addresses. But within a management domain, layers of routers generate excess complexity and require large staffs of “router guys.” They also bypass a primary tenet of many routing protocols—that every router should have a direct link to every other router that it knows about. The routing network is much more effective (and easier to manage) if all of the entities’ routers are directly interconnected, which is easily done today using optical Ethernet.
WAN
Ethernet transport has not yet taken off in the long-haul network, but this is expected to change as 10-Gigabit Ethernet interfaces become available. Some of those are expected to operate at SONET OC–192 speeds and at the distances needed for long-haul networks. The distance limitations are not a serious concern because most long-haul networks use dense wavelength division multiplexing (DWDM) systems to combine multiple circuits over a single fiber, each on its own wavelength, and these DWDM systems provide the long-haul capability themselves.
Still, wide-area Ethernet networks are expected to be implemented, the main reasons being speed, cost, and simplicity. Consider a nationwide 10-gigabit IP SONET–style ring implemented as OC–192 packet-over–SONET (POS) links between a dozen cities. Each city would need a large router with two relatively expensive POS ports, and the average packet would traverse half a dozen routers as it crossed the network. In the event of a link failure, the routers would spend a significant span of time converging on a new set of routing tables to bypass the failure. Now consider the same network with each city containing an Ethernet switch with two 10–Gigabit Ethernet switch ports and a 1-gigabit port connected to the local router. Total costs here would be much lower because 10–Gigabit Ethernet switch ports are expected to be much less expensive than the equivalent router ports, and 1-gigabit router ports are relatively inexpensive. Each router sees a direct connection to every other router, simplifying table lookups. And in the event of a link failure, the Ethernet switches themselves reroute the traffic more quickly at Layer 2—the routers do not need to be involved.
Today, several vendors have products that aggregate and transport Ethernet traffic at 10-gigabit speeds suitable for the WAN. These are proprietary, requiring matching devices from the same vendor at each end— normally not a problem in a full-duplex, point-to-point network.
Figure 3. LAN–CAN–MAN–WAN
