Martin Bech proved that “management stuff can be geeky” with a slide set of statistical equations used to calculate SLAs from actual availability figures. If you quote your actual mean downtime as an SLA this ensures that you will breach your SLA around half of the time. In fact measurements of the actual availability of unprotected links in the Danish NREN suggests that suppliers quote SLAs that are likely be breached only once in sixteen quarterly measurement periods. Thus, for example, a link with a SLA availability of 99.7% turns out to experience only 0.11% downtime on average, not 0.3%. When converting supplier SLAs into SLAs for connected sites, the same figure can be used if connectivity is dependent on a single link, but where a site depends on two network links in sequence the SLA should be lower (since a failure of either link will result in loss of connectivity); conversely if a site has two alternative routes to the network its SLA can be higher than the supplier quote (since the site will remain connected unless *both* links fail).
To calculate how much higher or lower, the SLAs need to be converted into mean availabilities (assuming that the distribution of failure lengths is exponential and using the 1 in 16 breach rate produces a simple conversion table), the mean availabilities combined to get a mean availability for the total connectivity, and this mean availability converted back into an SLA. For simple networking configurations the more accurate calculation does not make much difference but if different SLAs are offered or demanded it can provide important information to determine the appropriate network technologies and topology.