Remote Direct Memory Access (RDMA) can be used to implement a shared storage abstraction or a shared-nothing abstraction for distributed applications. We argue that the shared storage abstraction is overkill for loosely coupled applications and that the shared-nothing abstraction does not leverage all the benefits of RDMA. In this thesis, we propose an alternative abstraction for such applications using a shared-on-demand architecture, and present the RDMA Migration Platform (RAMP). RAMP is a lightweight coordination service for building loosely coupled distributed applications. This thesis describes the RAMP system, its programming model and operations, and evaluates the performance of RAMP using microbenchmarks. Furthermore, we illustrate RAMPs load balancing capabilities with a case study of a loosely coupled application that uses RAMP to balance a partition skew under load.