Data hotness and data density based cache back-writing method and system

Abstract

The invention relates to a data hotness and data density based cache back-writing method and system. The method comprises: S1, creating a least recently used (LRU) list array with n+1 elements for each cache of a disk array group; S2, calculating the data density of each cache block of the cache; S3, according to the data density, linking the cache block to an LRU list corresponding to an array subscript; and S4, when the cache block to be subjected to back-writing is selected, traversing the LRU list array in an inverted order, selecting the cache block with highest data density, and performing back-writing. With the method, relatively high cache back-writing efficiency is achieved when the hit rate of random write is medium.

Description

technical field [0001] The invention relates to disk storage technology, in particular to a cache write-back method and system based on data heat and data density. Background technique [0002] The least recently used LRU (Least Recently Used) algorithm is used to select the cache block that needs to be written back, because when the cache is large enough, the least recently used cache data block is the data with the lowest heat. This method has a good effect when performing sequential writing, because the earliest written data will not be written again, the heat is the lowest, and its data continuity, that is, the data density must be the highest. [0003] In the case of random writing and many parts hit the cache (for example, the random interval is 10GB, and the cache size is 5GB), the LRU algorithm has the following problems: (1) Since the data is very random, the average heat of all data is very low, so There is no difference between the write-back data selected at ran...

AI Technical Summary

Problems solved by technology

[0003] In the case of random writing and many parts hit the cache (for example, the random interval is 10GB, and the cache size is 5GB), the LRU algorithm has the following problems: (1) Since the data is very random, the average heat of all data is very low, so There is no difference between the write-back data selected at the time and the write-back data randomly selected, which degenerates into a random algorithm; (2) The data density is not guaranteed, and the write-back speed of the buffer block with higher data density is significantly faster than that of the buffer block with low data density

Images