Optimizing multi-homed connections: Enterprises can limit internet performance issues and hold their ISPs more accountable for service quality.

The Internet is coming of age, in large part because of its ability to open up markets, to make them more transparent. If I want to buy a new microwave, I can click on MySimon (www.mysimon.com), and instantly search dozens of Internet vendors and get a list of microwaves organized by price. The natural result of this process, however, is that most Internet-based offerings of the same product come in at virtually the exact same price.

Internet service providers (ISPs) will soon face similar pressures. Service providers like Inter-NAP constantly monitor ISP performance and availability. They report back to their enterprise customers on a regular basis and can switch traffic to the best available service. A new breed of route managers also has appeared, whose job is to help enterprises with multiple Internet connections (multihomed) optimize their connections (see David Passmore's columns in BCR, October and November 2001, pp. 20-22 and 18-20).

So What's The Problem?

Actually, there isn't so much a problem as an opportunity. Large customers, especially those who need constant Web presence are connecting to multiple carriers at each of their locations.

The most immediate benefit this provides is high availability. With a redundant Internet connection, the temporary failure of one service provider does not cut the Web connection. But, like many other redundancy situations, we don't want the second link standing idle while it waits for the occasional ISP failure. Instead, we want to use that bandwidth, and experience "graceful degradation" when one link goes down. In the meantime and, importantly, most of the time, we can put that bandwidth to work.

Bandwidth sharing is the first logical step. If I can manage my traffic and balance the load between links, no link becomes overutilized. This can improve overall performance by minimizing peak loading but, more importantly, it can help reduce cost by keeping traffic within contracted traffic agreements. With many contracts using "average" and "burst rate" charges, shifting traffic among Internet Service Providers to stay below surcharges is important. Management of traffic against bandwidth contracts based on traffic types, time of day and/or demand (load) provides real value, and can demonstrate a positive ROI for many businesses.

What About Performance?

Does it make sense to route traffic through one carrier or another to obtain better overall network performance? Users would certainly say yes.

Performance directly affects the user's perceived experience, which can positively affect a host of marketing issues--customer retention, return rate and conversion rate (see "Understanding Web Performance," in BCR, October 2001, pp. 28--36). A pack of new product vendors, discussed below, believe they can find the better performing route and deliver it dynamically.

Performance problems can be broken down into first mile, middle mile and last mile issues, as shown in Figure 1. Most folks are familiar with the last mile, where bandwidth is often a constraint. The last mile includes the access link and the ISP serving the client, and this part of the connection has only one path choice--e.g., one ISP and one …