By default, a pay-as-you-go DSW instance in a public resource group has a 100 GB system disk. When storage is insufficient, you can expand the system disk or mount a dataset. The two options differ in read/write speed, persistence, and data sharing, and suit different scenarios.
Quick comparison
Expand the system disk for temporary data and I/O-intensive workloads such as databases, log processing, and analytics. Mount a dataset for data that must persist or be shared, such as training datasets, model parameters, and result data.
| Item | System disk expansion | Dataset mounting |
| Read/write speed | Fast | Slower. Speed varies by storage type: OSS, NAS, or CPFS |
| Ease of expansion | High. Simple operation with no service interruption | Low. Requires additional configuration |
| Persistence | Low. Content is reclaimed after the DSW instance is deleted | High. Data is stored in cloud storage such as OSS, NAS, or CPFS |
| Data sharing | Not supported | Supported. You can share data across multiple instances and cloud services |
| Data security | Relatively low. Data is cleared when the DSW instance is deleted | High. Data is stored durably in cloud storage such as OSS, NAS, or CPFS |
| Capacity reduction | Not supported. The capacity cannot be reduced after expansion | - |
| Billing while the instance is stopped | You continue to be billed for the expanded capacity | - |
| Applicable scenarios | I/O-intensive workloads and temporary storage | Persistent storage, shared access, and data security |
Option 1: Expand the system disk
Advantages
Fast read/write speed. The system disk is a high-performance storage device with fast read/write speeds, suitable for I/O-intensive workloads.
Easy expansion. You can expand the system disk through a simple configuration change. After expansion, data on the system disk is retained even if the disk remains unused for a long time. The data is reclaimed only when you delete the instance.
Disadvantages
Inconvenient data sharing. The system disk is bound to a single instance. Data cannot be shared across instances.
Non-persistent storage. System disk content is retained after the instance is stopped, but is cleared when the DSW instance is deleted.
Irreversible. The capacity cannot be reduced after expansion.
When you stop a pay-as-you-go DSW instance, compute resource billing stops. However, you continue to be billed for the expanded system disk capacity because it still occupies cloud disk resources. To stop all charges, back up your data and delete the instance.
Use cases
Temporary storage. Suitable for scenarios that require quick data access.
I/O-intensive workloads. Suitable for applications that require high read/write speeds, such as databases, log processing, and analytics.
Option 2: Mount a dataset
Advantages
Persistent storage. Data is independent of the DSW instance lifecycle and is stored in datasets in cloud storage such as OSS, NAS, or CPFS. Data is not lost even if the instance is shut down or deleted.
Data sharing. Share data with other DSW instances or cloud services. Multiple instances can read the same dataset data through the mount path.
Data security. Dataset storage is more reliable and secure than system disk storage.
Disadvantages
Slower read/write speed. Accessing data in a dataset may be slower than accessing data on the system disk.
Additional configuration required. Mounting a dataset requires additional configuration and management and imposes certain technical requirements. See Mount a dataset, OSS, NAS, or CPFS.
Use cases
Persistent storage. Suitable for data that needs to be stored long-term, such as training datasets, model parameters, and result data.
Shared access. Suitable for scenarios where multiple instances or users access the same dataset, such as collaborative projects and team workflows.
Data security. Suitable for applications that require high data persistence and security.
Next steps
For instructions to mount and manage datasets, see Mount a dataset, OSS, NAS, or CPFS.
Upload and download files: For files larger than 5 GB, upload the files to an OSS bucket, create an OSS dataset, and mount it to a DSW instance to read OSS data directly.