Optimizing a canvas app, especially one with slow initial launch and hybrid data sources, requires a multi-faceted approach.
1. Moving data calls out of the App.OnStart property is a critical first step. OnStart is a blocking function; the app will not render until all its operations complete. By deferring data calls to a screen's OnVisible property (the likely intent of the term Onload in the question), the app can load its user interface first, improving perceived performance.
2. Reducing the amount of data retrieved from sources is a fundamental performance principle. Loading excessive rows slows down the app and increases memory usage. Employing delegation and filtering to retrieve only necessary data, and setting appropriate data row limits, directly improves responsiveness.
3. For on-premises data, the data gateway is a potential bottleneck. Its performance depends on network latency and the resources of the host machine. Reviewing gateway logs, monitoring performance counters, and ensuring the host server is adequately provisioned are essential troubleshooting steps.

C. Increase data row limits. This would retrieve more data, further degrading app performance, which is the opposite of the required optimization.
D. Move data calls from the Onload event to the OnStart event. This would make the initial launch performance worse by moving data operations into the blocking OnStart function.

1. Microsoft Learn. (2023). Tips and best practices to improve canvas app performance.

Reference for A: In the section "Use App.OnStart sparingly," the documentation states, "To improve the performance of an app, move formulas from App.OnStart to other properties in the app... For example, you can move some formulas to the OnVisible property of the first screen." This directly supports moving data calls out of OnStart.

Reference for B: In the section "Data sources," the documentation advises to "Limit the number of columns retrieved" and "Use delegation." This supports the principle of reducing the amount of data requested to improve performance.

2. Microsoft Learn. (2023). On-premises data gateway monitoring and optimization.

Reference for E: This document details how to troubleshoot slow-performing queries that use a gateway. It recommends actions such as enabling additional logging, checking network latency, and ensuring the gateway machine is not resource-constrained (CPU, memory, networking). This directly supports reviewing gateway settings for performance issues.

3. Microsoft Learn. (2023). Understand delegation in a canvas app.

Reference for B: This document explains that non-delegable queries process data locally after retrieving the first 500-2000 records (the data row limit). It states, "The key is to keep your data row limit low so that your users' expectations are clear." This reinforces that managing and reducing data retrieval is key to performance.