AnalyticDB for MySQL elastic clusters utilize an architecture that separates storage resources from computing resources.

In addition to data storage, AnalyticDB for MySQL storage nodes also provide data writing, data searching, and data scanning capabilities. By default, the number of storage I/O resources is one third of that of computing resources. If the read/write performance reaches the upper limit, you must scale out elastic I/O resources in the AnalyticDB for MySQL console.

An elastic I/O unit (EIU) is the unit of data that measures the storage performance of AnalyticDB for MySQL elastic clusters.

The following table describes storage performance metrics of each EIU.

Metric Maximum value
Hot data storage space 4 TB
IOPS 16,800 to 50,000
Throughput 350 MB/s
CPU 24 cores

In query scenarios where data is written with high throughput and a large amount of data is scanned, EIU resources may have I/O and CPU bottlenecks. You can scale out EIUs to expand related capacities and scale in EIUs when business workloads decrease. You can use similar methods to scale out and scale in EIUs.

You can use the following method to scale out EIUs:

  1. Add an EIU node.
  2. Migrate part of shards from the original node to the new node based on the load balancing algorithm.
    • The original node takes snapshots of shards to be migrated and transfers the snapshots to the new node.
    • The shards to be migrated on the original node are still online and provide services during migration.
    • The new node loads the snapshots and starts to append shard logs.
    • When the shard logs on the new node are the same as those on the original node, the new node starts to receive new writes and queries.
  3. The shards that are migrated are gracefully stopped on the original node. After existing queries end, these shards are offline and cleared.

EIU scale-out has the following features:

  • EIU resources can be scaled out and scaled in at any time based on business requirements.
  • EIU scale-out or scale-in does not affect your business. Clusters do not need to be stopped, and data reads and writes are not affected.