Virtual network (VNet) data gateways help organizations connect Microsoft Fabric workloads to data sources that are protected by virtual network boundaries. They are commonly used when teams need private connectivity for semantic models, Dataflow Gen2, and DirectQuery scenarios without changing their existing connectivity architecture.
The modern evaluation engine is now available in preview as an opt-in setting for VNet data gateways. This feature introduces an updated execution path for supported workloads while continuing to use the same private network gateway model. The Modern evaluation engine applies only to Mashup evaluations that are executed through a VNet data gateway.
Because the feature is currently in preview, organizations should evaluate it using representative workloads before broader deployment. Customers can test the updated execution path without changing their gateway deployment model or network architecture. During preview, the setting is off by default. Gateway administrators can enable it for selected gateways, validate representative workloads, and decide when to expand usage based on their own testing.
Why this matters
Many organizations rely on VNet data gateways to provide private connectivity between Microsoft Fabric and business-critical data sources. As adoption of semantic models, Dataflow Gen2, DirectQuery, and other analytical workloads grows, improving execution efficiency can have a meaningful impact on day-to-day operations.
The modern evaluation engine focuses on helping customers:
- Evaluate refresh performance for supported workloads.
- Reduce execution overhead in selected scenarios.
- Evaluate DirectQuery cold-start behavior.
- Adopt improvements without changing existing gateway architectures.
- Evaluate benefits through a simple opt-in experience.
In internal benchmark scenarios using specific datasets and configurations, supported semantic model refresh, Dataflow Gen2 refresh, and DirectQuery cold-start workloads completed faster when using the modern evaluation engine. Actual performance depends on workload design, connector behavior, source-system performance, gateway configuration, and workload concurrency. Because every environment is unique, customers should validate performance using workloads that are representative of their production environment.
Who should evaluate the feature?
The preview is particularly relevant for organizations that:
- Use VNet data gateways extensively today.
- Depend on large semantic model refreshes.
- Run Dataflow Gen2 workloads through private network connectivity.
- Use DirectQuery with cold-start scenarios against protected data sources.
- Need to maintain strict network isolation requirements.
- Want to evaluate potential execution improvements without redesigning existing architecture.
Customers already using VNet gateways are likely to find the evaluation process straightforward because the feature builds on existing gateway deployments.
What should you measure?
To get the most value from the preview, consider monitoring the following metrics before and after enabling the feature:
- Refresh duration
- Track total execution time for semantic model refreshes and Dataflow Gen2 refreshes. Comparing historical averages against new results helps determine whether the modern evaluation engine provides measurable improvement for your environment.
- Query responsiveness
- For DirectQuery scenarios, monitor cold-start and steady-state query behavior using representative reports and datasets.
- Reliability
- Validate that refreshes, queries, and scheduled workloads behave consistently throughout testing.
- Capacity utilization
- Review Fabric capacity metrics to understand how workload execution characteristics change over time. By measuring multiple dimensions rather than focusing solely on elapsed time, organizations can make more informed rollout decisions.
Get started
To try the modern evaluation engine:
- Open Manage connections and gateways.
- Select a VNet data gateway.
- Turn on Use modern evaluation engine. The following setting enables the modern evaluation engine on a selected VNet data gateway:
Figure: Enable the Use modern evaluation engine setting on a VNet data gateway to evaluate supported workloads using the updated execution path.
4. Run representative semantic model refreshes, Dataflow Gen2 refreshes, and DirectQuery workloads.
5. Compare performance and capacity metrics against your baseline.
Because the capability is opt-in during preview, administrators retain full control over where and when it is evaluated.
Best practices during preview
To maximize the value of testing:
- Start with a limited rollout.
- Test representative workloads rather than synthetic scenarios only.
- Compare against documented baselines.
- Monitor performance over multiple execution cycles.
- Include both refresh and query scenarios in testing.
- Review capacity metrics throughout the evaluation period.
- Expand adoption gradually as confidence increases.
These practices help ensure that observed results accurately reflect expected production behavior.
Learn more
The modern evaluation engine enables organizations to evaluate an updated execution path for supported VNet gateway workloads while continuing to use their existing private connectivity architecture.
If you use VNet data gateways for semantic models, Dataflow Gen2, or DirectQuery workloads, this preview offers an opportunity to assess potential improvements using your own workloads and operational metrics.
Next step: Enable Use modern evaluation engine on a non-production VNet data gateway, run representative workloads, compare the results against your baseline, and review the documentation to determine whether the feature is a good fit for your environment.
Learn more in the Modern evaluation engine for VNet data gateways documentation.