In this work the connection availabilities of optical transport networks restorable to singlespan failures have been investigated. Consideringthe fast increase of data carricd by networks and the recent emphasis on hardcning them against natural disasters and terrorist activities, enhancing the robustness of the networks has become a priority for most Service Providers. However, the traditional approach of availability analyses is to determine the end-to-end availability just for a few selected reference connections, which is no longer a good representative of network Performance and cannot be used to develop specific design recommendations for availability and network optimization. In addition, viable methods and exaet results about connectionavailabilities were providedonly for unprotected or dedicatedly protected networks. The analyses for distinet restoration schemes were condueted only from a general point of view of availability and have to rely on completely different and complicated approaches. A unified new concept of availability analysis is proposed in this dissertation and a practical method is developed to determine connectionavailabilities in a mesh-based survivable network with various survivability schemes by assuming dual-span failures to be the only failure mode and an arbitrary allocation rule of spare capacity. According to this concept, basic computation modeis are built and validated both by simulations and by analytical approaches for the four most promising survivabilityschemes at present, i.e. dedicatedautomatic protection switching, shared backuppath protection, span restoration and pre-configured protection cycles. The correspondingcomputational funetions are concluded for each scheine. Thus connectionavailabilities of all possible demand pairs can be exactly computed not only for networking using the simple dedicated protection scheine, but also for mesh-restorablenetworks with shared spare capacity, for which only a rough estimate of the availability Performance could be made previously. The influence of various factors is then analyzed on connection availabilities and the average dual-span failure restorability. A network designed with a higher redundancy achieves better dual-span failure restorability, and always a small increase in redundancy can lead to a big restorabilitygrowth. Since the total investment is usually required to be kept as low as possible, several capacity designs of a span-restorable network have been presented in this work to maximize the average network dual-failure restorability under a given budget. This can assist network planncrs to optimize the dual-span failure restorabilityagainst investment costs. Someunavailability-relatedmeasures have been developed, e.g. averagecxpected downtime or traffic loss per year of all connections for an assessment of the absolute network Performance and some composite measures for a Performance comparison of networks with different survivability schemes, like the ratio of the average connection availability to the network redundancy. Most of the Performance measures are of the same order of magnitude, though the scheme of shared backup path protection shows the best Performance with the most efficient capacity design. The quantitative connection availability, computed from the proposed new concept, can definitely help Service providers to set availability guaranteesmore exactly and safely in service level agreements (SLAs). Finally a theoretical basis is presented, by assuming different distributions of failures and repair times, to set "safety factors" of the availability guarantees for various contract periods in SLAs, which, in practice, are just based on empirical experience. It is shown that improving the overall system availability is not only a laudable end in itself, but makes it also less likely that an SLA guarantee is not fulfilled even with short contract periods.