Different modules of a multi-threaded application are tightly coupled to each other due to shared data structures, global contexts and dependencies between threads. This tight coupling prevents the application designer from distributing different parts of the application across machine boundaries. Even if the designer decides to distribute, she must use distributed algorithms and primitives for synchronizing large distributed applications like distributed shared memory, commit protocols, distributed locking, election algorithms etc. Several middle wares in academia and industry have tried to provide such primitives to the application programmer. However, in all such systems, the programmer has to be aware of the distribution and must use them judiciously to her advantage. The issues of dealing with complexity of usage of such primitives mentioned for single machine systems described earlier increase by an order of magnitude for highly distributed systems.