The diffusion model based task remapping for distributed real-time systems
Morikazu Takegaki, Hiroo Kanamaru, M. Fujita · 2002
A fault-tolerant architecture for loosely coupled distributed real-time systems is considered and a distributed task management method that aims at network-wide fault tolerance, compatible with flexible resource utilization is proposed. The task management method includes a checkpoint data transfer mechanism and an adaptive task remapping mechanism. A policy of task remapping is based on the thermodynamic diffusion model, in which tasks are diffused among sound nodes and global balance is achieved by several repetitions of local load sharing. Convergence to this steady state is theoretically ensured. The protocol of the proposed task remapping algorithm, the two-phase diffusion algorithm, is completely distributed and weakly synchronized; each node starts the same algorithm periodically, without waiting for any messages, but waits for messages from neighboring nodes during execution of the algorithm.>