I recall having calculated several years back that ~4 hours of
unavailability (to customers) per 30-day month implies 99.44% availability,
an Ivory tower number to be sure (are there any of you out there old enough
to remember the incessant ad's for Ivory Soap "99 and 44/100 percent
Sprint proposed such a number to NSF back in 1989 for the international
services under the International Connections Management (ICM) agreement.
One thing we both learned along the way was that there are factors beyond
any carrier's abiolity to control, especially in international (read
"multi-player") services. So, for example, when a cisco box needed
re-powering in Stockholm or London, but the local staff was gone for the
night and nothing would happen until morning, or when the physical circuit
passed through Amsterdam or Germany, and a line break depended on PTT
Netherlands or Deutsche Telekom to repair, could/should we hold Sprint
responsible? We decided to ask Sprint to produce numbers that related to
the portion of the end-to-end linnk under their direct control, e.g., ICM
box in Washington to ICM box at London demarc. The numbers, of course,
looked a lot better. (I cannot recall them anymore.) But even then, we
did not impose quality measures like packet drop rates (and, if we had, how
could we know if they were caused by traffic congestion or by poor line
quality if the link was heavily loaded almost all day?).
Just to mudy the waters a bot more...