Data-driven adaptive checkpoint optimization method

Abstract

The invention discloses a data-driven self-adaptive check point optimization method, which comprises the following steps of: calculating CPU node failure distribution by taking fault data in a fault history library as a sample and adopting a maximum likelihood estimation method; establishing an application failure distribution model according to the CPU node failure distribution; calculating to obtain an optimal check point interval according to the failure distribution model; guiding the system to check points at an optimal check point interval; when a new fault occurs, updating related CPU node failure distribution; jumping to step 2. Through the real fault data and the adaptive distribution model optimization algorithm, the operation-level failure distribution model and parameters thereof are dynamically updated, the operation check point interval is continuously optimized and set, the check point interval is adaptively and dynamically adjusted, the check point fault-tolerant modelis optimized and perfected, the check points are optimized as much as possible, and the fault-tolerant performance of the system is improved. Therefore, the fault-tolerant overhead based on check points is reduced, and the availability level of the system is greatly improved.

Description

technical field [0001] The invention relates to a data-driven self-adaptive checkpoint optimization method, which belongs to the technical field of computers. Background technique [0002] In any fault-tolerant system using checkpoint or restart technology, it is necessary to strike a balance between checkpoint overhead and computational overhead in order to achieve fault tolerance with the lowest overhead. Too frequent or too long checkpoint intervals will lead to an increase in system fault tolerance overhead and lost calculations. An optimal checkpoint strategy can minimize checkpoint overhead and improve fault tolerance efficiency. [0003] At present, the checkpoint system adopts a fixed-interval checkpoint mechanism, and the time interval is generally set by manually specifying the experience value. This method is not combined with real system reliability, too frequent or too sparse checkpoints will cause unnecessary checkpoint overhead or increase the amount of lost ...

AI Technical Summary

Problems solved by technology

This method is not combined with the real system reliability. Too frequent or too sparse checkpoints will cause unnecessary checkpoint overhead or increase the amount of lost calculations.

Images