All Products
Search
Document Center

Platform For AI:System disk expansion and dataset mounting

Last Updated:Apr 01, 2026

By default, a Data Science Workshop (DSW) instance created in a pay-as-you-go public resource group comes with a 100 GB system disk. When that space runs out, you have two options: expand the system disk, or mount a dataset.

Quick rule: Expand the system disk for high-speed temporary storage. Mount a dataset when data must persist after an instance is deleted, or when multiple instances need access to the same data.

Quick comparison

FeatureSystem disk expansionDataset mounting
Read and write speedHighLow. Speed depends on the backing storage type: Object Storage Service (OSS), File Storage NAS (NAS), or Cloud Parallel File Storage (CPFS).
Ease of expansionHigh. Update the configuration — no service interruption.Requires additional configuration and technical operations.
PersistenceLow. The disk is reclaimed when the DSW instance is deleted.High. Data lives in cloud storage (OSS, NAS, or CPFS) independently of any instance.
Data sharingNot supported. Attached to a single instance.Supported. Multiple instances and cloud services can access data through the mount path.
Data securityLow.High. Data is stored durably in OSS, NAS, or CPFS.
DownscalingNot supported. Expansion is irreversible.N/A
Billing when instance is stoppedStorage billing continues for expanded capacity.Billed by the backing storage service.

What to store where

Use the following table to match your data type to the right storage option.

Data typeRecommended optionReason
Training datasets (especially > 5 GB)Mount a datasetPersists across instance deletions; shareable across instances
Model weights and checkpointsMount a datasetMust survive instance stops and deletions
Shared team dataMount a datasetMultiple instances can read the same data through the mount path
Intermediate files during a training runSystem disk expansionHigh I/O speed; data does not need to outlive the session
In-memory databases, real-time log ingestionSystem disk expansionRequires low-latency block storage access
Code and scriptsSystem disk expansionAccessed frequently; does not need cross-instance sharing

System disk expansion

The system disk is block storage attached directly to the instance. This gives it low latency and high throughput, but data exists only for the lifetime of the instance.

Advantages

  • High I/O performance. Direct block storage access makes system disks suitable for workloads that need fast read/write speeds, such as databases, log processing, and real-time data analysis.

  • Simple to expand. Update the disk configuration — no restart or service interruption required. The expanded disk remains available until you delete the instance, even if the instance sits idle for an extended period.

Limitations

  • No cross-instance sharing. The disk is attached to one instance and cannot be accessed by others.

  • Data lost on instance deletion. Data is retained when an instance is stopped. All data is permanently deleted when the instance is deleted.

  • Irreversible. Once expanded, the disk capacity cannot be reduced.

Important

Billing for expanded system disk storage continues after you stop the instance — only compute resource billing stops. To stop all charges, back up your data and delete the instance.

Dataset mounting

Datasets mount cloud storage (OSS, NAS, or CPFS) to a path inside your DSW instance. Because data lives in the cloud storage service — not on the instance — it persists independently of the instance lifecycle.

Advantages

  • Persistent storage. Data survives instance stops and deletions. Training datasets, model weights, and result data remain intact regardless of what happens to the instance.

  • Cross-instance sharing. Multiple DSW instances or other cloud services can access the same data simultaneously through the mount path.

  • High durability. OSS, NAS, and CPFS provide reliable and secure storage that is independent of any single instance.

Limitations

  • Lower read and write speed. Compared with system disks, the speed of accessing data in a dataset may be lower.

  • Additional setup required. Mounting a dataset involves specific configuration steps. See Mount a dataset, OSS, NAS, or CPFS.

What's next