A communication-induced checkpointing and asynchronous recovery algorithm for multithreaded distributed systems

Checkpointing and recovery in traditional distributed systems is relatively well established. However, checkpointing and recovery in multithreaded distributed systems has not been studied in the literature. Using the traditional checkpointing and recovery algorithms in multithreaded systems leads to false causality problem and high checkpointing overhead. The checkpointing algorithm is implemented at the process level to reduce number of checkpoints and the recovery algorithm is implemented at the thread level which minimizes the false causality problem. The algorithm also takes advantage of the communication-induced checkpointing method to reduce the message overhead.