In this paper, we present two availability constrained adaptive connection provisioning techniques that work under dual failure assumption. The proposed techniques are called, G-DAP (global differenciated availability-aware provisioning) and LBL-DAP (link-by-link differenciated availability-aware provisioning) respectively. As a heuristic approach, G-DAP attempts to determine a feasible sharing degree for each availability class on the backup channels. LBL-DAP uses an optimization model to determine a feasible sharing degree for each availability class on each optical link separately. The performance of these adaptive techniques are compared with a conventional reliable connection provisioning algorithm, CAFES (compute-a-feasible-solution) where connection requests arrive with the availability requirements 98%, 99%, 99.9%, 99.99% and 99.999% under NSFNET topology. Based on the results obtained, it is shown that the presented techniques lead to higher acceptance rate and higher availability per connection. Moreover, LBL-DAP also leads to the lowest resource overbuild. Therefore, LBL-DAP can be considered as the best performing reliable connection provisioning technique out of the presented techniques here.