The distribution of services among nodes can be highly customized. Veritone provides the following general guidelines.
Sizing aiWARE nodes
For multi-node aiWARE, one of the nodes must be the controller node. If you are installing AI Data, then allocate a node for AI Data. Other nodes are processing nodes. All nodes must run containers.
Controller nodes
- Docker or Mirantis containers
Processing nodes
For processing nodes that will run engines, ensure that you account for the following.
- Docker or Mirantis containers
- 32GB of swap space
- Root partition of at least 100 GB
- 4 AI Data servers or more
- For the Docker root directory of an engine node, a minimum of 500GB disk space
For processing nodes that run AI Data services, account for the following.
- Docker or Mirantis containers
- Large disk space, optimized for I/O with low latency storage For the node that has the registry run mode, allocate 500GB of disk space for images needed to run engines.
For all aiWARE instance configurations, ensure that you have the minimum services for aiWARE, as in the following table.
| aiWARE Configuration | Minimum services in one or more nodes | Possible distribution of services in multi-nodes |
|---|
| AI Processing | controller, db, engine, nfs | Controller node: controller, nfs, db Processing node: engine |
| AI Data | controller, db, engine, nfs, lb, minio, nsq, api, es | Controller node: controller, nsq, db
Processing node: engine
AI Data node: api, apps, lb, minio, es |
