Interval availability estimation for protected connections in optical networks

Routing and wavelength assignment under availability constraints has been extensively researched recently. Availability (or more precisely, steady-state availability) can be defined as the average probability of a connection operating over a time window that tends to infinity. However, service level agreements (SLAs) commit a minimum connection uptime fraction over a finite contract duration. This random variable is known in reliability engineering as interval availability. If the minimum agreed interval availability is not honored, the service provider is penalized. In order to balance the risk of non-compliance fines against asset protection costs, network planners must know the interval availability distribution. However, its estimation with existing numerical techniques is computationally expensive, motivating the search for approximate analytical methods. Under the hypotheses of Poissonian node and link failures and repairs, and assuming no more than two link failures or one node failure in the network, we propose, for connections protected by shared or dedicated methods:
• an approximate Markov model that allows the derivation of a closed-form expression for the connection steady-state availability;
• under the approximate Markov model, analytical bounds on the interval availability distribution.The proposed methods are validated by discrete-event simulations of an Italian network.