FAULT TOLERANCE IN THE MOBILE ENVIRONMENT

Abstract

In general it is assumed that a parallel program will execute on reliable hardware. A fault tolerant program and underlying infrastructure should be capable of surviving fail- ures such as system crashes and network failures. At the highest level the application should be capable of automatically recovering from a set of faults without any change to the apparent behaviour of the program. The process of checkpointing may be used to allow a program to save its state to persistent storage, abort and restart from the checkpoint. Several fault tolerant MPI implementations are currently in existence, for example MPICH-V is considered to be one of the most complete, featuring checkpointing and message logs to allow aborted processes to be replaced. No matter how sophisti- cated a fault tolerant system may be, it can never be completely relied upon, as there is always the possibility of a complete system failure. It is one thing to develop fault toler- ant applications on high end dedicated clusters and supercomputers, however applying fault tolerance to the realm of mobile parallel computing presents an entire new series of challenges that are inexorably linked with the unpredictable nature of wireless com- munication systems. Two differing strategies for fault tolerance in the mobile Bluetooth wireless environment will be presented and compared to see which should be adopted over another.