Currently, only RDS for SQL Server 2017 Enterprise Edition supports the Cluster Edition (AlwaysOn Edition), which is based on the native AlwaysOn technology of SQL Server. The Cluster Edition separates computing from storage, and supports read/write splitting. By default, each read-only instance in the cluster also provides an independent intranet connection string.

The following figure shows the architecture of the Cluster Edition.

Advantages

  • Horizontally scalable read capability
    You can add read-only instances to linearly increase the cluster's read capability. Additionally, the specifications of read-only instances can be different from those of the master instance. Therefore, you can add read-only instances with high specifications to achieve even better read capability.
    Note By default, read-only instances do not support high availability. To implement high availability of read-only instances, at least two read-only instances are required.
  • Flexibly controllable costs

    In the Cluster Edition, you can create read-only instances of common specifications, which are cost-effective options. Additional read-only instances can process more read requests so that the overall system configuration is optimized.

    In addition, specifications of read-only instances can be lower than those of the master instance. Therefore, you can choose cost-effective read-only instance with lower specifications for background applications, such as intelligent analysis applications.

    In the near future, the Cluster Edition will also support the maximum performance mode, which can be used during peak hours to implement asynchronous replication between the master and slave nodes, so that the cluster performance is maximized.

Scenarios

  • Use read-only instances to handle read requests during peak traffic hours

    For example, to prepare for peak trading events like Double 11 Festival, online retail enterprises can buy additional read-only instances with high specifications to process the majority of read requests while implementing read/write splitting and traffic control on the service level. In this way, the enterprises can handle several times the amount of usual traffic.

  • Assign analysis tasks to read-only instances

    Enterprises generally have demand for intelligent data analysis. An independent read-only instance dedicated for data analysis can reduce the probability of the master instance becoming unresponsive, improve concurrency, and reduce the impact on core service queries, thereby ensuring service stability.