The triplicate storage technology for Alibaba Cloud disks uses a distributed file system to provide stable, efficient, and reliable random data access for Elastic Compute Service (ECS) instances. This ensures the data reliability of your ECS instances. This topic describes the concept and principles of triplicate storage.
Introduction to triplicate storage
Alibaba Cloud provides a flat, linear storage space that is internally sliced into Chunks. Each Chunk is replicated into three replicas. These replicas are stored on different data nodes in a storage cluster based on a specific policy to ensure data reliability. Your read and write operations on a disk are ultimately mapped to read and write operations on files on the Alibaba Cloud data storage platform.
For data on a disk, all user-level read and write operations are synchronized across the three underlying replicas. This triplicate model ensures the reliability and consistency of your data.
How triplicate storage works
The Alibaba Cloud data storage platform has three roles: Master, Chunk Server, and Client. A write operation is ultimately executed by the Client. The process is as follows:
The Client receives a write request and determines the corresponding Chunk for the operation.
The Client queries the Master to find the data nodes (Chunk Servers) that store the three replicas of the Chunk.
Based on the response from the Master, the Client sends a write request to these three Chunk Servers.
If the write operation succeeds for all three replicas, the Client returns success. Otherwise, the Client returns failure.
To prevent data from becoming unavailable due to the failure of a Chunk Server or a rack, the Master ensures that the three replicas are distributed on different Chunk Servers across different racks. Therefore, the Master's distribution policy considers various factors on the data storage platform, such as the disk usage of all Chunk Servers, the distribution of switches, power supply conditions, and node workloads.
Data protection mechanism
If a data node is damaged or some disks on a data node fail, the number of valid replicas for some Chunks in the cluster might fall below three. In this case, the Master initiates an automatic synchronization task to copy data between Chunk Servers. This ensures that the number of valid replicas for every Chunk in the cluster is restored to three.
If data is lost from an ECS instance due to soft failures, such as a virus infection, accidental deletion, or a hacker attack, use technical measures such as backups or snapshots to recover the data. No single technology can solve all problems. Choose appropriate data protection measures for your specific needs to build a strong defense for your business data. For more information about snapshots, see Snapshot overview.