This topic describes the resource management schemes that are supported by ApsaraDB for SelectDB to help you better use resources in ApsaraDB for SelectDB.
Resource management schemes
ApsaraDB for SelectDB supports the following schemes that you can use to manage and isolate computing resources: multi-instance, multi-cluster, and workload group.
Multi-instance scheme
ApsaraDB for SelectDB allows you to create multiple independently deployed instances within the same account. This scheme is often used in scenarios in which systems are completely independent of each other.
In this scheme, resources are isolated based on instances. The computing and storage resources of an instance are physically isolated from those of another instance. Each instance has its own endpoints.
Multi-cluster scheme
ApsaraDB for SelectDB supports the storage-computing separation architecture. You can create multiple physically isolated compute clusters within the same instance. Each compute cluster consists of a group of backend (BE) nodes. A computer cluster is similar to a compute group or compute queue in a big data system. This scheme is often used in scenarios in which you want to isolate read traffic from write traffic or isolate online traffic from offline traffic.
In this scheme, computing resources are isolated based on clusters. However, the clusters within an instance share the same set of data. You can connect to multiple clusters in an instance by using the same endpoint and run the USE @cluster_name command to switch to the cluster that you want to use. For more information, see Multiple compute clusters.
Workload group scheme
ApsaraDB for SelectDB allows you to use workload groups to isolate computing resources in a software manner. A cluster can have multiple workload groups. This allows you to restrict the usage of computing and memory resources on a single BE node by each task. This scheme is often used in scenarios that require resource isolation but allow resource preemption to a certain extent.
In this scheme, computing resources are logically isolated based on workload groups. However, all workload groups share the same set of data. You can use the database accounts that are associated with workload groups to access different workload groups. For more information, see Workload group.
Scheme comparison
The following table describes the comparison of the preceding resource management schemes in terms of isolation and resource utilization.
Scheme | Isolation method | Isolation level |
Multi-instance scheme | Computing resources and data of different instances are completely isolated from each other. | Ultra-high |
Single-instance multi-cluster scheme | Computing resources are physically isolated, but data is shared among compute clusters. | High |
Workload group scheme | Computing resources are logically isolated, but data is shared among workload groups. | Medium |
The multi-instance scheme excels other schemes in isolation but may result in redundant storage resources and low utilization of computing resources. In contrast, the workload group scheme supports moderate preemption of computing resources. This ensures high utilization of resources but provides weak isolation. In general, we recommend that you use the multi-instance scheme in scenarios in which your systems are completely independent of each other. In scenarios in which data needs to be shared but computing resources must be physically isolated, we recommend that you use the multi-cluster scheme. In scenarios in which data needs to be shared and computing resources sometimes need to be isolated, we recommend that you use the workload group scheme.
Scheme recommendation
You can combine the preceding schemes based on your business requirements. For example, you can combine the multi-instance scheme with the multi-cluster scheme or combine the multi-cluster scheme with the workload group scheme.
Multi-instance scheme
You can consider using the multi-instance scheme in the following scenarios:
No data is shared between independent business systems or teams.
Some business systems are important or contain sensitive information, and you must minimize the probability of data leaks and prevent mutual interference.
Multi-cluster scheme
You can consider using the single-instance multi-cluster scheme in the following scenarios:
Data is shared, but computing resources need to be isolated for data with different sensitivity levels. For example, the computing resources for online data analysis need to be isolated from those for offline data analysis.
The workloads for data import need to be isolated from those for data queries.
You have no special requirements on resource utilization.
Workload group scheme
You can consider using the workload group scheme in the following scenarios:
Data needs to be shared, and computing resources do not need to be physically isolated for data with different sensitivity levels. Basic resources must be available for each type of data, and idle resources can be preempted.
You want to make full use of resources or have special requirements on resource utilization.