Automate financial risk simulation runs with .NET

After you validate a small run, automate the workflow by using the current Azure SDK libraries. Use Azure.Compute.Batch for jobs and tasks, Azure.ResourceManager.Batch for pools and account resources, Azure.Storage.Blobs for input and output data, and Azure.Identity for authentication.

Use the Batch .NET quickstart to set up the clients and run a validated sample. Apply the following simulation-specific design to your application.

Create one traceable job per run

Create a unique job ID and store the run manifest with the job outputs. Add only identifiers to Batch metadata; keep portfolio data and other sensitive values in protected storage.

If a run has many partitions, submit tasks in collections of up to 100. Give each task a deterministic partition ID and associate only the resource files that partition needs.

Persist and aggregate outputs

Configure task output files to upload results and diagnostics to Azure Storage. Use a hierarchy such as <run-id>/<task-id>/<attempt> so that retries don't overwrite unrelated results.

Enable task dependencies on the job if a final task aggregates the partition outputs. The aggregation step should verify the partition count and output schema before it publishes the final result. An external orchestrator can perform the same validation after all tasks complete.

Design for re-execution

Batch can retry or requeue tasks after application errors, node failures, pool resizing, or Spot node preemption. Make tasks idempotent so that running a partition more than once produces one valid result.

• Use deterministic partition identifiers and seed inputs.
• Write attempts to separate paths and publish only validated results.
• Set finite retry and maximum wall-clock limits.
• Checkpoint long calculations to durable storage when the model supports restart.

For configuration guidance, see Design for retries and re-execution and Error handling and detection.

Authenticate with Microsoft Entra ID

Use Microsoft Entra ID credentials such as DefaultAzureCredential for the submitting application. Assign a user-assigned managed identity to the pool when compute nodes need access to Azure Storage or Azure Container Registry. Grant only the required data-plane roles. For more information, see Configure managed identities in Batch pools.

Use a job schedule or an external workflow orchestrator for recurring runs after the submission and validation logic is reliable.

Next step

Evaluate GPU acceleration for financial models