TS-PVM: a fault tolerant PVM extension for real time applications

In this research work, a fault tolerant extension of the de facto message passing system parallel virtual machine, TSparallel virtual machine, is introduced. This extension enables real time applications over parallel virtual machine. In PVM and similar message passing systems, if the message receiver is not available, due to network failure or machine crash, a failure is reported and data must be resent. If the transferred data is persistent, i.e. should not be lost, then it is important to provide the system with failure recovery mechanisms. The idea behind the proposed extension, TSparallel virtual machine, is to integrate a fault tolerant distributed shared memory layer, in parallel virtual machine. Such layer has been inherited from the TRIPS system that supports fault tolerance over a DSM. The challenge of this research work is the integration of a DSM layer in a message passing system. The proposed parallel virtual machine extension would have bad performance if compared to the pure message passing one. On the other hand, the real time application can be completed in spite of failures occurrence.