A Dynamic Grouping System Based on MPI High Performance Computing Hierarchical Rollback Recovery Protocol

Abstract

The invention discloses a dynamic group system for layered rollback recovery protocols based on MPI (Message Passing Interface) high performance computing. The system comprises a message monitoring module, a message analysis module and a process migration module, and belongs to the fields of high performance computing and system fault tolerance. The message monitoring module is used for monitoring a message passing record between processes in an MPI high performance computing application, storing the record in a format of a triad (source process, target process and message size) and submitting the message passing record to the message analysis module; the message analysis module is used for analyzing the message passing record collected by the message monitoring module, analyzing to obtain a message passing mode of the current application as a basis for the application to execute the process migration module at the next time while utilizing the last message passing mode to judge whether to execute the process migration module at this time; the process migration module is used for carrying out migration operation on the changed process when the message passing mode of the application changes, so that the performance optimization of the layered rollback recovery protocols is realized.

Description

technical field [0001] The invention belongs to the field of high-performance computing and system fault tolerance, and more specifically relates to a dynamic grouping system based on MPI high-performance computing layered rollback recovery protocol. Background technique [0002] With the development of the field of high-performance computing, high-performance computers have grown to a scale of one million nodes, and there may be further growth in the future. At the same time, the mean time between failures (Mean Time Between Failures, MTBF) of high-performance computer systems has dropped significantly compared with before, even reaching the order of several hours. However, the data size, computational complexity, and running time of distributed commercial applications and large-scale scientific computing applications remain at a high level, and even the running time lasts for several months, which is much greater than the MTBF. This will cause the system to spend too much...

AI Technical Summary

Problems solved by technology

[0005] For the above defects or improvement needs of the prior art, the purpose of the present invention is to provide a dynamic grouping system based on the MPI high-performance computing layered wrapping recovery protocol, aiming at solving the inability of the grouping mechanism of the existing layered wrapping recovery protocol Adapting to technical inefficiencies caused by changes in application communication patterns

Images