Triplicate technology

Last Updated: Jun 08, 2017

The Alibaba Cloud Distributed File System provides stable, efficient, and reliable random data access capabilities for ECS.


When ECS users perform read and write operations onto virtual disks, the operations are translated into corresponding processes on the files stored in the Alibaba Cloud Distributed File System. Alibaba Cloud provides a flat storage space, in which the linear addresses are divided into chunks, also referred to as slices. Alibaba Cloud employs a certain strategy to create three copies for each chunk and stores these copies on different nodes, ensuring the reliability of user data.

Principles of triplicate technology

The Alibaba Cloud data storage system consists of three roles: Master, Chunk Server, and Client. The write operation of an ECS user goes through several conversions and is executed by the Client. The procedure is as follows:

  1. The Client calculates the chunk corresponding to a given writing operation.
  2. The Client sends a request to the Master for the storage location of the three copies of the chunk.
  3. The Client sends writing requests to the three Chunk Servers according to the results returned from the Master.
  4. The Client returns a message to the user indicating whether the operation was successful.

The distribution strategy of the Master is decided based on an overall consideration of the following:

  • Disk use conditions of all Chunk Servers in the cluster.
  • Distribution of the Chunk Servers under different kinds of switch racks.
  • The power supply.
  • The instrument load.

This strategy ensures that all the copies of a Chunk are distributed on different Chunk Servers on different racks. This can effectively prevent data unavailability caused by the failure of a Chunk Server or rack.

Data protection mechanism

When some data nodes are corrupted, or some hard drives on a certain data node fail, the number of valid copies of some Chunks in the cluster will be less than three. If this occurs, the Master initiates the copy mechanism to copy data between Chunk Servers, making three valid copies of all Chunks in the cluster.

In sum, for the data on the cloud disk, all user operations and data addition or modification will be synchronized to the three copies. This mode ensures the reliability and consistency of user data.

To prevent data losses caused by virus infection or cyber-attacks, we recommend that you use the triplicate technology with other protection methods, such as taking snapshots.

Thank you! We've received your feedback.