I/O requests area forecasting method based on sequence-degree clustering algorithm and time series

Abstract

The invention discloses an I / O area prefetching method based on continuity clustering and time series. According to the feature of locality of the I / O request, the invention predicts the area of ​​intensive I / O access when the system is idle, and prefetches the predicted area. It uses a clustering algorithm based on continuity, which can efficiently and reliably discover areas of intensive read requests; secondly, it uses the ARMA time series model to predict the areas and access times that future intensive read requests may visit. Under the same test environment, the storage system of the present invention and the existing RAID system have been compared and tested, and the load test using 3 disk volumes shows that: the clustering algorithm of the present invention can correctly cluster read requests; simultaneously based on AMRA Accurate prefetch-intensive I / O areas for dynamic forecasting algorithms for time series forecasting models. The invention can practically and efficiently discover the areas accessed by intensive read requests, accurately prefetch the areas that may be accessed by intensive read requests, and greatly improve the performance of the storage system.

Description

technical field [0001] The invention relates to the field of storage, in particular to an I / O area prefetching method based on continuity clustering and time series. Background technique [0002] Although the parallel I / O system (disk array) can greatly improve the performance of the storage system, there is still a large performance gap between them due to the inherent defects of the disk (longer seek delay and rotation delay). . As a basic implementation technology, caching can make up the performance gap between them very well, but simply increasing the capacity of the cache is not an effective solution to the problem. The prefetching technology predicts future data requests and stores The data blocks in the device are fetched into the cache before they are used, so that the request can always be hit in the cache to reduce the CPU pause time, and the contention for the disk or channel is eliminated by predicting the cache miss and overlapping I / O technology, reducing the...

AI Technical Summary

Problems solved by technology

However, in the existing file system, the storage system at the device layer does not know any semantic information of I / O access, so it cannot make full use of the semantic information of I / O access to prefetch the data to be accessed at the next moment. Use simpler methods such as locality of I / O access, sequential access, and iterative access to achieve simple predictions

At the same time, we know that in I / O-intensive applications, I / O accesses have bursty characteristics, and we can prefetch the next read request-intensive time at a time when I / O access is less (system idle time) All the data to improve the performance of the system, the large-capacity cache in the existing storage system also makes this method possible

Images