Resource configuration
API resource configuration
Intelligent Workshop lets you specify resource configurations in API calls to meet various business requirements for performance, stability, and cost. If you do not specify a resource configuration, the system uses shared computing power resources by default.
Resource types
1. Shared computing power resources (default)
Definition: A system-level shared resource pool used by multiple users and workflows.
Scenarios:
Workflow development and debugging.
Non-critical tasks that are not sensitive to execution latency.
Billing method:
Time-based billing: Billing starts only when a task starts executing.
Pay-as-you-go model: Charges are settled after tasks are executed.
Resource scheduling:
Tasks are scheduled in a queue, which may cause waiting times.
Compute nodes are shared among multiple tenants. This can cause resource interference between tasks, such as when GPU memory is not fully released.
Potential risks:
Increased risk of GPU memory overflow, performance fluctuations, and task failures.
2. Dedicated computing power resources (recommended)
Definition: Computing resources reserved for a single user. This guarantees dedicated computing power and stable performance.
Scenarios:
Business scenarios that require low execution latency and a high task success rate.
Online production tasks and sensitive inference or generation workflows.
Configuration method:
Configure and reserve a dedicated resource pool in Resource Management.
You can reserve resources by time or configure them based on resource usage.
Usage:
Specify the dedicated resource configuration ID as a parameter in the API call.
Advantages:
Dedicated scheduling with no queueing.
Higher task stability and a lower risk of resource interference.
Controllable costs, suitable for long-running critical tasks.
Other
Recommendations
Business type | Recommended resource type | Reason |
Workflow debugging | Shared computing power resources | Low cost and meets basic operational needs. |
High-availability production tasks | Dedicated computing power resources | Stable performance and high task success rate. |
Concurrent multitasking scenarios | Dedicated computing power resources | Avoids queueing and improves processing efficiency. |
Cost-sensitive businesses | Shared computing power resources | No resource reservation required. Billing is based on actual usage. |
FAQ
Q: What happens if I do not configure a resource ID? A: The shared resource pool is used. This can cause queueing and performance fluctuations. This option is not recommended for services that require high stability.
Q: Am I always billed for dedicated resources? A: If you choose the "reserve by time" pattern, the resources are billed continuously. If you use the "by resource usage" pattern, resources are scaled out and billed on demand only when a usage threshold is met.
Q: How do I request dedicated resources? A: In the Resource Management module, you can submit a resource request, configure a resource group, and then generate a resource configuration ID.