SLAs, who needs them?

Remember when "SLA" triggered thoughts of Patty Hearst? Today, networking people think of AT&T, MCI and Sprint when they hear the acronym, which has become a hot topic of discussion following last month's AT&T frame relay outage.

One of the more common support arguments for SLAs is that they make network outages easier to handle financially. For example, AT&T agreed to pay customers a monetary reimbursement for the duration of the outage. This is great thinking if you're trying to run your IS department as a cash-positive operation, but it may not be the best use of, or argument for, SLAs. Here's why.

Let's take a trip back into the not-so-distant past when everything in the wide area was circuit switched, and packet switching was just a dream in the minds of a few lab engineers and marketing managers. When users purchased a leased-line circuit, they received 100% of the bandwidth. If the circuit failed, they would rely on their backup connection - typically an equal or lower speed line in place for just such a contingency. In some cases, this was even a dial-up connection (perhaps an inverse T-1 multiplexer).

In this relatively simple model, service levels were fairly easy to determine. The circuit was either up or it was down. When it was up, you had 100% of the bandwidth with a nice fixed latency. And when it failed, you had a contingency plan (and perhaps a rebate plan from your provider to compensate for the "down" time of the primary circuit).

Fast-forward to the late 1990s. Frames and cells are everywhere. A circuit can still be up or down, but "up" is now a relative term. Today, there are many different shades of gray when it comes to a service being active or not. Latency can vary considerably, as can bandwidth utilization. Applications may or may not work.

To counter this variability in our networks, many users have turned to SLAs as a mechanism for establishing a set of parameters for an "active" network connection - often linking a certain performance level (much like a Frame Relay CIR) to a specific charge. If the service levels asked for by the user are not met, then the users pays a pro-rated fee.

Unfortunately, many managers have begun using SLAs as a mechanism for determining monetary compensation only and are even willing to put up with degraded service if they can essentially get a discount on it. This kind of thinking, however, can become deadly in an age when QoS-dependent applications are increasingly coming online and when business-critical functions rely more and more on packet-switched networks.

In this type of network, 90% bandwidth availability, or an increase in latency by 10%, may be just as devastating as a total circuit failure. Thus, I would argue that you should use SLAs not to determine the cost of a particular level of service, but to determine at what threshold to demand the activation of backup service when the main connections degrade.

Rather than demanding that service providers deliver discounts or rebates for certain levels of poor service, perhaps the better course is to ask for that old leased line - 100% bandwidth utilization - instead. For example, when service falls below 98% availability, beyond a certain millisecond delay, or below a certain bandwidth usage, the carrier agrees to cut the user's connection over to a fixed, dedicated line.

It may not be popular with the service providers, but it sure will work.