Checkpointing and recovery for mobile distributed systems: a new approach

摘要

Traditional checkpointing and rollback recovery schemes designed for a static network, are not suitable for applications running on a mobile computing system (3, 11). Issues like mobility of the nodes, long disconnection periods of mobile hosts (MHs), lack of stable storage in MH, high fault rate and available limited communication bandwidth of wireless links needs to be considered. Traditional checkpointing and recovery schemes for static networks can be broadly classified into synchronous, asynchronous and quasi-synchronous [5]. All these strategies have been explored for designing checkpointing schemes for mobile environment. This paper presents a checkpointing technique which uses probabilistic approach to avoid excessive checkpointing activity at mobile nodes in "no receive after send (NRAS)"[1, 3] checkpointing algorithm, which is SZPF protocol[5]. In our scheme a mobile node skips some of the checkpoints which may be needed to construct a consistent global state in case of a fault. Our protocol uses the message logs to construct these skipped checkpoints if they are needed during recovery. We show that the probability that a skipped checkpoint will be required during recovery is very low at moderate fault rates. We also present a simple recovery algorithm for distributed mobile systems with this checkpointing activity. The probability to take a checkpoint at mobile node depends on the fault rate and message pattern. This probability makes the system adaptive to the fault rate and message pattern, and reduces the load on the wireless links. The proposed scheme is compared with the existing deterministic schemes through simulation.