Fast parallel recomputing method aiming at data parallel computing fault tolerance

Abstract

The invention belongs to the technical field of parallel system fault tolerance, relates to a parallel recomputing method using redundancy computation for error detection and error correction on a computing task, and particularly provides a fast parallel recomputing method for logic division and secondary scheduling on the basis of a data block corresponding to an error task. The method comprises the following steps of: performing error detection based on a redundant computing strategy on the computing result of the data block; and performing recomputing on the basis of multi-thread error detection and process recomputing error correction. The method provided by the invention can be completely applied to fault-tolerance processing occasions of high-performance calculation of parallel digital terrain analysis of large-scale mass data, such as the terrain factor extraction of rule grid parallel interpolation, slope and aspect parallel computing, depression filling and leveling-up parallel computing and the like; the method can be applied to high-performance computing of geographic information processing, and can also be applied to the application occasions of space decision analysis based on geographic information, data mining and the like; and the processing efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of fault tolerance of parallel computing systems, relates to using redundant strategies to quickly correct errors in computing tasks, and in particular proposes a fast parallel recalculation method for data parallel computing fault tolerance. Background technique [0002] Fault-tolerant processing of parallel computer systems is a problem that cannot be ignored. A parallel system is fault-tolerant, which means that its program can still run correctly and guarantee correct results even in the event of logical failures. [0003] In recent years, with the increasing complexity of computer system structure, the development of semiconductor manufacturing process, the reduction of line width and the improvement of integration, from user desktop systems to distributed computing environments, and even large-scale parallel computer systems, power consumption and reliability Sexual issues are becoming increasingly pr...

AI Technical Summary

Problems solved by technology

The current mainstream software fault-tolerant strategy is oriented to time redundancy, which leads to the failure of the node to restore the task again, because its recovery time is greater than the time interval between the previous checkpoint and the moment of failure, resulting in tasks with dependencies Waiting for a long time, and these problems lead to reduced parallel efficiency and waste of computing resources

Images