This topic describes the servers required for deploying OceanBase AI Database.
OceanBase AI Database consists of two components, OBServer and LogService, and currently supports two deployment modes: separate deployment and mixed deployment.
- Separate deployment: OBServer and LogService are deployed on different machines, and the two components scale independently.
- Mixed deployment: OBServer and LogService are deployed on the same machine, and the two components can still scale independently.
Before preparing the servers, you need to choose an appropriate deployment plan based on your business requirements so that you can determine the required machine resources. For a detailed introduction to deployment plans, see Choose an appropriate deployment plan.
Server configuration requirements
The server requirements are as listed in the following table.
Product/Component |
Number of servers |
Minimum configuration requirements |
Disk Type |
|---|---|---|---|
| OAT | 1 server (can reuse the OCP server) | CPU: 2-core Memory: 4 GB |
N/A |
| OCP and MetaDB | 1 server | CPU: 8-core Memory: 24 GB Disk: 200 GB storage, 10 Gbit/s NIC (including resources required for OAT) 10 Gbit/s NIC (including resources required for OAT) |
SSD Storage |
| LogService | 3 servers | Calculated by using the following formula: Cache disk: 100 GB |
SSD Storage |
| OBServer | 1 or more servers (can reuse LogService node servers) | CPU: 4 cores Memory: 16 GB Cached disk: Recommended to be more than 3 times the memory size. If there is hot data, it is recommended to be 1.5 times the amount of hot data. 10 Gbit/s NIC Note
|
SSD Storage |
| (Optional) ODP | 3 servers (can reuse OBServer node servers) | CPU: 4-core Memory: 8 GB Disk: 200 GB |
N/A |
The resource calculation formula for deploying LogService is as follows:
Assume that the number of OBServer replicas is R, the number of units per replica is U, the number of CPU cores per node is C, and the number of LogService nodes within a zone is N. Then the resources of a LogService node should satisfy the following formulas:
- Number of CPU cores P: CEILING (1 + ((0.1 + (R - 1) * 0.05) * U * C) / N, 0.5)
- Memory M (GB): CEILING(P * 3, 1)
- Local disk (GB): 100
- Local disk bandwidth (MB/s): (P + 1) * 100
Example:
If the number of OBServer replicas is 2, the number of nodes per replica is 2, the number of CPU cores per node is 32, and the number of LogService nodes in a zone is 1, then R=2, U=2, C=32, N=1. The recommended resources for a single LogService node are:
- Number of CPU cores: CEILING(1 + ((0.1 + (2 - 1) * 0.05) * 2 * 32) / 1, 0.5)=11
- Memory (GB): CEILING(11 * 3, 1)=33
- Local disk bandwidth (MB/s): (11 + 1) * 100=1200
Note
- If you need the OCP management service to provide high availability, you must deploy multiple management servers for high availability. For details, see OCP Deployment Overview.
- For detailed host resource requirements of OCP and MetaDB, see Host Resources. For service resources, see CPU and Memory.
- The high availability of an OceanBase AI Database cluster is guaranteed by LogService. LogService must be deployed in 3 replicas, requiring at least 3 nodes. Depending on disaster recovery requirements, OBServer can be deployed in 1 replica or 2 replicas.
- To ensure LogService resource usage, cgroup-based resource isolation must be used to isolate CPU, memory, I/O, and local disk bandwidth resources.
- Whether ODP is required depends mainly on your application scenario and requirements:
- If your application requires high availability, load balancing, or connection pool management for the OceanBase AI Database cluster, you may consider deploying ODP. ODP can effectively address connection and load-balancing issues and improve application stability and reliability.
- If your application only needs to connect to the OceanBase AI Database cluster for queries and operations, and does not require high availability, load balancing, or connection pool management, you do not need to deploy ODP. You can connect directly to the cluster using the client libraries provided by OceanBase.