Applications /

An SLA is not a substitute for engineering quality service into your network

By Joel Snyder
Network World on Groupware and Messaging, 09/22/99

I recently wrote an opinion piece in Network World about service-level agreements. Some readers wrote in asking me to expand my comments. So here's a longer version about what you should know about SLAs.

Let's start with the basics. The whole concept of an SLA is flawed. It says that a company (the seller) agrees to provide a certain level of service to another company, and if the seller doesn't do what it promises, the seller has to fix it.

That's all well and good, and an SLA can be useful for a buyer to know what's being bought. But the SLA is really little more than a description of the service being bought. If I say "I want 20 ms ping time coast-to-coast," it doesn't matter whether the Purchase Order says "Frame relay service, 20 ms round trip latency" or "Frame Relay Service, see attached SLA," it's the same thing.

The concept is the same with outsourced electronic mail ("POP connects and authentication within 1/2 second") or any service-oriented product.

Service providers are using fancy new terminology to somehow justify the lack of any service description in the past. Before, they never said what you bought: you just got it. Now, they say what it is. So in that sense SLAs are nothing more than the normal due diligence that any purchasing agents go through when they buy a service for their companies. For service providers to trumpet this as a triumph, or competitive edge, makes them look stupid, at least to me. Agreeing to write down what you just sold someone doesn't sound to me like an innovative approach.

Besides, an SLA really doesn't do anything. Just because I have an SLA that says I'm buying such-and-such a service doesn't mean that my provider is going to deliver it. All it says is that if they don't, they fix it or I don't pay. I see nothing new and innovative about this. Normally, when someone doesn't deliver a product, I don't feel obligated to pay for it. More importantly, in an SLA, I see nothing that guarantees a business that the crucial services it needs are going to be there when there's a problem.

As an example, look at the recent MCI WorldCom frame relay outage. Sure, MCI WorldCom is giving some companies some money back. But the value of crucial services that MCI WorldCom customers lost due to the frame relay outage is much higher than what those customers are paying dollarwise to MCI WorldCom. Just look at those ISPs and companies that depended on MCI WorldCom's service for everyday business.

Probably most MCI WorldCom customers that were affected by the frame relay outage didn't have an SLA. But the effect on a company's business was the same, with or without the SLA. The difference is merely that without an SLA, MCI WorldCom gets to decide what it's going to give the customer as compensation.

I am not blaming MCI here, though. The real problem is that companies have gotten cheap and are engineering their networks closer and closer to the bone. The Internet is a great example: Why do you think it's so inexpensive? Because large parts of it are subject to massive single points of failure. Any chunk of the Internet can go off the air at any time with a single slipped keystroke. That it works as well as it does is a testament to the quality of the people who run it and the gear that they use. But as a network itself, it's tremendously vulnerable. Connect your company to the Internet and save thousands of dollars on leased lines. But you're hanging out there in the breeze if anything goes wrong.

There is only one way to ensure quality service, and that's to engineer it into your network yourself. You can pick good partners, and good equipment, but the bottom line is that service providers will go down and will fail to provide the service they've promised. Having a piece of paper with "SLA" printed on the top isn't going to change that. Having a network that is fault tolerant is.