A kind of intelligent write allocation method and device based on zns solid state hard disk

Abstract

The present invention provides an intelligent write allocation method and device based on a ZNS solid-state hard disk. The method includes: receiving data to be written issued by a storage application; allocating a target write object for the data to be written; extracting the target write object The multi-dimensional data attribute feature is mapped, and the multi-dimensional data attribute feature is mapped to the write partition type; according to the write partition type, a write area is allocated for the to-be-written data. The intelligent write allocation method and device provided by the embodiments of the present invention can classify and store write data through object attribute feature extraction oriented to storage applications, so that objects of different cluster types can be stored separately, sequential and random objects can be stored separately, and cold storage can be maintained. The hot data objects are stored separately, which reduces the performance loss caused by the mixed data storage, avoids the classification failure caused by the loss of real data attributes based on the LBA data classification algorithm, and maintains the generality of the algorithm, while giving full play to the performance advantages of flash media.

Description

technical field [0001] The invention relates to the field of ZNS solid-state hard disks, in particular to an intelligent write allocation method and device based on ZNS solid-state hard disks. Background technique [0002] High-performance SSD solid-state drives are widely used in enterprise storage systems as a substitute for disks. SSDs follow the access interface of disks and shield the operation characteristics of flash memory. In the prior art, storage applications convert logical addresses into LBA addresses. The internal FTL converts the LBA logical address to the actual physical address of the NAND. When data is written, the data of different modes and temperatures are mixed and stored, which cannot give full play to the performance advantages of the flash memory medium. Since the NAND needs to be updated in different places, the logical data is overwritten. The resulting invalid data requires garbage collection, and data mixing will seriously affect garbage collecti...

AI Technical Summary

Problems solved by technology

[0002] High-performance SSD solid-state drives are widely used in enterprise-level storage systems as a substitute for disks. SSDs follow the access interface of disks and shield the operating characteristics of flash memory. In the prior art, storage applications convert logical addresses into LBA addresses. SSDs The internal FTL converts LBA logical addresses into NAND actual physical addresses. When data is written, data of different modes and temperatures are mixed and stored, which cannot give full play to the performance advantages of flash media. Since NAND needs out-of-place updates, logical data overwrites The resulting invalid data needs garbage collection, and the performance of garbage collection will be seriously affected when the data is mixed. Therefore, the introduction of write classification technology in the solid-state storage system can optimize the performance of the overall storage system

[0003] The current writing classification technology is mainly divided into two types. One is to classify and store data for a specific storage application. This classification technology is not versatile and cannot adapt to different storage applications; the other is The SSD internally classifies and stores data at the LBA level. Since the LBA address received by the SSD loses the application layer attribute information, the LBA converted by the storage application logic is already different from the original data attribute. Two fixed-size files Write them sequentially, and two data streams cross cycle overwrite storage as an example. When the disk is empty, the LBA addresses of the data written by the SSD are in complete order and belong to different files. When the storage space is full, you need to delete the old file and generate a new one. The LBA occupied by the deleted file itself is not continuous when the file is written and deleted, causing the subsequent LBA rules of the new file generated after recycling to be disrupted. Therefore, the data classification technology based on the LBA cannot effectively extract the true attributes of the data.

Images