A Tunable Checkpointing Algorithm for the Distributed Mobile Environment
Sungchae Lim · 2011
The aim of a distributed checkpointing algorithm is to efficiently restore the execution state of distributed applications in face of hardware or software failures. Originally, such algorithms were devised for fixed networking systems, of which computing components communicate with each other via wired networks. Therefore, those algorithms usually suffer from heavy networking costs coming from frequent data transits over wireless networks, if they are used in the wireless computing environment. In this paper, to reduce usage of wireless communications, our checkpointing algorithm allows the distributed mobile application to tune the level of its checkpointing strictness. The strictness is defined by the maximum rollback distance (MRD) that says how many recent local checkpoints can be rolled back in the worst case. Since our algorithm have more flexibility in checkpointing schedule due to the use of MRD, it is possible to reduce the number of enforced local checkpointing. In particular, the amount of data transited on wirelesses networks becomes much smaller than in earlier methods; thus, our algorithm can provide less communication cost and shortened blocking time.