Multi-Model Fault-Tolerant Programming in Distributed Object-Based Systems

Extended and replicated transaction models provide consistency and forward progress guarantees for distributed applications that manipulate persistent, shared data. In our earlier work, we developed OS mechanisms that can implement the various transaction models as policies at the user level. The mechanisms facilitate eeciency and exibility, but programming using the primitive OS mechanisms could be complex. To simplify the programmer's task, in this paper, we develop higher level abstractions. These abstractions can be implemented using the underlying system mechanisms, as part of a language run-time system and library routines. We informally argue the exibility of our approach for programming transaction models that facilitate nested and colored actions, and Parallel Execution Threads (PET). We quantify the costs of fault tolerance through a prototype implementation of multiple transaction models on the Clouds distributed system.