Data storage method, device and system

Abstract

The invention discloses a data storage method, device and system. The data storage method comprises the following steps of carrying out erasure code coding on a data block to obtain a plurality of corresponding data fragments; respectively writing the plurality of data fragments into the correspondingly distributed storage nodes; if the number of the storage nodes successfully writing the data fragments in the first set duration is greater than or equal to a set value, determining that the data block is successfully written, wherein the set value is greater than or equal to the number of the data block fragments in the data fragments and less than or equal to the number of the data fragments. Therefore, the data writing delay can be reduced, and the performance stability when the storage node has a fault can be improved.

Description

technical field [0001] The present application relates to but not limited to the technical field of data storage, and in particular relates to a data storage method, device and system. Background technique [0002] Current distributed storage systems increasingly use Erasure Coding (EC, Erasure Coding) technology to store data. Taking Reed-Solomon (RS, Reed-Solomon) type EC as an example, its technical principle is: divide the data into K data blocks, use redundancy algorithm to check and encode the K data blocks, and generate M checks blocks; wherein, when less than or equal to M blocks among the K+M blocks are lost, the data in the lost blocks can be restored. In order to ensure the reliability of storage, the distributed storage system will hand over K+M blocks to K+M storage nodes for storage. In this way, even if a node fails, data recovery can be performed based on data stored in other nodes. [0003] The implementation of EC in the current distributed storage syste...

AI Technical Summary

Problems solved by technology

However, in the above-mentioned direct writing (Direct) EC scenario, if some storage nodes involved in the writing process are abnormal, so that it cannot be guaranteed that all storage nodes are successfully written, the write will be retried until the timeout or If the write fails, rebuild the data on the storage node that timed out or failed to write, and then continue to write after the reconstruction is completed; in this way, a relatively large write delay and write failure rate will be generated, resulting in a relatively large fluctuation of the direct write EC delay. Large, unstable performance

Images