A Fault-Tolerant Method for Checkpoint-Based Computers

Abstract

The invention provides a fault tolerance method based on a check point for a computer. The method comprises the following steps: when a judgment of executing the check point is made, suspending a user process; segmenting process state data of the user process by using idle computing resources in the computer, and computing a hash value of each block in order to determine blocks needing to be saved; and saving the determined blocks needing to be saved and the corresponding hash values while computing the hash values of the blocks in order to form a check point file for recovering a wrong user process. Through adoption of the method provided by the invention, the idle computing resources in a super computer and an I / O (Input / Output) bandwidth of a parallel file system can be utilized effectively, so that check point execution time and check point rollback time are shortened.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a fault-tolerant method for computers (especially supercomputers) based on checkpoints. Background technique [0002] With the development of information technology, the number of nodes and processors of supercomputers continues to increase, and the performance is also doubled. However, according to statistics, the mean time between failures (MTBF, Mean TimeBetween Failure) of the entire supercomputer system has been reduced to Only a few hours. For example, the Tianhe-2 supercomputer in China consists of 16,000 nodes, each with 2 Ivy Bridge-E Xeon E5 2692-based processors and 3 Xeon Phi co-processors, for a total of 32,000 Ivy Bridge processors and 48,000 Xeon Phi co-processors, with a total of 3.12 million computing cores. If the MTBF of each processor in the Tianhe-2 supercomputer is 876000 hours (100 years), then the MTBF of the entire Tianhe-2 is 876000 / (48000+32...

AI Technical Summary

Problems solved by technology

This method uses the page size as the block size of the process state data. This method will cause the checkpoint to determine the granularity of the process state data modification content, and the checkpoint based on the page protection mechanism requires the support of the operating system and hardware. The checkpoint Using the copy-on-write technology of the operating system will also cause the parent and child processes to rob computing resources and memory resources

This method does not utilize the idle computing resources during the execution of the checkpoint process, and does not effectively utilize the I / O bandwidth of the parallel file system, and cannot satisfy the I / O of the parallel file system that quickly rolls back and reduces rollback in the event of an error bandwidth requirements

Images