Parallel runtime error detection method

Abstract

The invention provides a method for detecting errors during parallel operation, comprising: setting a first counter and a second counter with an initial value of 0; when a process enters an MPI blocking operation, adding one to the first counter and starting a timer; When returning from this blocking operation, the value of the first counter is assigned to the second counter, and the timer is canceled; and, if the MPI is blocked in an MPI call, a soft interrupt signal is triggered when the timer is full, thereby Enter an interrupt processing function, compare the current value of the first counter and the second counter in the interrupt processing function, if the current value of the first counter and the second counter is not equal, then perform a state dump and then perform deadlock detection; if If the current values ​​of the first counter and the second counter are equal, then return from the interrupt processing function, and continue to execute the parallel program.

Description

technical field [0001] The invention relates to the technical field of computers, in particular to a method for detecting errors during parallel operation. Background technique [0002] In the field of HPC (High Performance Computing, high-performance computing), MPI (Message Passing Interface, message passing interface) is currently the most widely used and flexible parallel programming model, and it is also the mainstream message passing standard, and the most widely used parallel program programming model, there are multiple open source implementations. For MPI, because it is not automatically parallelized, programmers generally develop parallel computing by gradually increasing the parallel scale from small to large. [0003] When the program is parallel in a small scale, you can add printing to the program or use the debugging method of the debugging tool to solve the detection and location of errors during parallel operation. However, when faced with hundreds of point...

AI Technical Summary

Problems solved by technology

[0008] The technical problem to be solved by the present invention is to provide a method for detecting errors during parallel operation in view of the above-mentioned defects in the prior art, which is oriented to MPI communication deadlock detection, and solves the problem that the prior art has a large operating cost and lacks parallelism during operation. The defect of interactive fault exploration and analysis hardly affects the performance of parallel programs when it is not activated. It can implement the runtime library based on the open interface of the MPI library. It has good portability and is completely transparent to user programming.

Images