There are a number of issues here. As mentioned by Vodacom3G is the challenge that SP's have to define what 99.9% actually means, and what they can guarantee when a number of factors are outside of their control. The other point is how do you measure the success or failure of the SLA.
For example, I may be an SP who hosts websites on a shared platform. All I can 'guarantee' is the availability of that platform. Although I can mitigate the risk of Telkom line failure, I cannot guarantee against it. All I can do is have back to back OLA's with Telkom so that if I end up paying penalties, then I get it back from them. (I'm not sure if Telkom would ever/have ever agreed to such arrangements).
What about if my customer updates their website and screws it up. They may see it as platform unavailability, when it is again something outside of my control as the hosting provider. Who's responsibility does it become to prove what happened?
What you end up with is the hosting provider having to put so many terms and conditions to their SLA that there will always be a loophole so that they are not responsible.
The other issue around how you measure the success or failure of the "99.9%" availability is also important. If a hosting provider said to me they could offer 99.9% availability, I would say "so what". The site may be available 100% of the time (i.e. you can 'ping' the web server) but that doesn't help me if the platform is so oversubscribed that the performance of may website makes it unusable - it's still 'available' but in practical terms may as not be. And then in an issue like this, is it because of the platform infrastructure performance or poor web site design - and again, who needs to prove that.
The solution is to first of all list the factors in and out of their control. See where back to back OLA's can be established to cater for the 'out of control' factors. Once you have decided on factors that you can base SLA's on, you need to see what you can measure. Depending on what you want to do, there are technologies available to monitor "End User Experience", server performance, application performance on servers, application performance on networks and more. What is key to this is that the monitoring should also provide sufficient analysis of problems to 'prove' responsibility.
In short, what is in my control? what is out of my control? Where can I set up OLA's? What can I measure? what can I prove?
This is a challenging subject and not something that can easily be addresses in a single forum response. For a hosting provider, service guarantees are difficult, but can be done a lot better than they are at the moment.