What you need to know to build a solid Ethernet WAN

Sourcing solid network connectivity, such as for linking a pair of data centers, isn't always as straightforward as it might seem

Building a resilient, high-capacity WAN has never been what I'd call simple. In the old days (you know, 10 years ago), it was typically a mishmash of frame-relay and point-to-point leased line circuits mixed with ATM in higher-end, converged applications. Today, those technologies and their successors are being displaced by a variety of IP-based Ethernet circuits. Although I think this shift to Ethernet-based WAN implementations is hugely liberating when it comes to WAN design, it has its drawbacks.

Chief among those drawbacks: It becomes a lot more difficult to know exactly what you're buying. When I buy a point-to-point T1, I know with some certainty that I'll have a 1.5Mbps channel from site A to site B. The data I push into the pipe from the router at site A will end up at the site B side in exactly the same order and typically with a very predictable and consistent latency.

Ethernet-based systems, while far more flexible and feature-rich, have a much higher degree of variability due in large part to the same flexibility they offer customers. As a result, you need to have a much better understanding of the technologies that modern Ethernet-based carriers use and what might differentiate the options you have available. Without that understanding, you might end up paying too much for a service that won't meet your needs in the long run.

The last mile comes first, and fiber is the best last mile

Perhaps the most crucial part of any WAN circuit is how the so-called last mile is delivered to you. In North America, there are several common ways used by carriers to deliver Ethernet-based services to your doorstep. The most important distinction is whether the last mile is based on copper (such as twisted-pair phone lines or the coaxial cable used in cable TV plants) or fiber optics.

Put bluntly, copper almost always loses in a reliability and scalability race against fiber. The main reason that copper still exists as a digital delivery mechanism is that some providers -- those with cable TV charters and the incumbent telephone carriers -- own enormous copper-based cabling infrastructures and want to suck every dime of revenue out of them while they can. Thus, network vendors have put a lot of elbow grease into developing fairly innovative ways of using those copper infrastructures to deliver next-generation networking services.

The real challenge with copper is that its bandwidth is limited. For example, the predominant way of providing symmetrical Ethernet services over PSTN twisted-pair copper is SHDSL. Using SHDSL, a provider can combine pairs of copper to provide as much as 50Mbps of symmetrical throughput. However, this maximum throughput figure falls off dramatically the further away from the central office your site is. For sites that need relatively small amounts of bandwidth (10Mbps and less), copper can be a great and inexpensive option. But if the site requires a lot more bandwidth (especially more than 40Mbps to 50Mbps), copper is almost sure not to stand the test of time.

1 2 Page 1
Page 1 of 2