In this thesis, we present the variants of two representative non-blocking linearizable protocols tailored to geo-distributed environments and discuss the trade-offs between them in terms of cost and latency. This thesis also presents a cost-effective linearizable geo-distributed key-value store called GeoStore. GeoStore takes as an input the properties of the workload being considered and predicts a suitable protocol and its configuration by modeling it as a quadratic optimization problem.It also shows that for certain workloads, erasure coding based linearizable protocols perform significantly better in terms of cost as compared to replication based protocol. As a future direction, this thesis also implements the optimistic version of these protocols which can further optimize them for certain workloads.