A system and method for checkpoint recording and recovery in a distributed environment

Abstract

The invention discloses a system and method for checkpoint recording and recovery in a distributed environment. The system includes three modules, namely a monitoring module, a checkpoint recording module and a checkpoint recovery module. The monitoring module is responsible for monitoring whether the process is running correctly, closing the recording module and starting the recovery module when the operation is abnormal; the checkpoint recording module is responsible for recording checkpoints in units of messages; the checkpoint recovery module is responsible for restoring inspection according to certain rules point; the monitoring module is respectively connected with the checkpoint recording module and the checkpoint restoration module, and the monitoring module, the checkpoint recording module and the checkpoint restoration module respectively monitor, record and restore the parent process operation information. Methods include: checkpoint recording and checkpoint recovery. The invention solves the problem that the distributed checkpoint cannot finally find the consistency checkpoint, the communication volume between processes is very small, and has the advantage of non-blocking of the distributed checkpoint.

Description

technical field [0001] The invention belongs to the field of computer science and relates to the reliability of computer clusters, more specifically, it is an asynchronous checkpoint technology protocol suitable for distributed environments and can be used for computer cluster error recovery. Background technique [0002] Currently, computing tasks are becoming increasingly complex and require ever-increasing computation times. At the same time, high-performance computing systems include an increasing number of failure-prone components. The end result is that long-running distributed computations are increasingly interrupted by frequent hardware errors. In distributed computing, when a process fails, the cost is not only the loss of all calculations of the process, but also the loss of calculations of the calculation processes communicating with it. In order to ensure that distributed applications can be used more effectively in large-scale environments, it is imperative t...

AI Technical Summary

Problems solved by technology

However, since the checkpoint is recorded randomly by the process itself, there may be a problem that a consistent state cannot be found

Images