Intelligent disk pre-staging for personal host pools

The intelligent disk pre-staging feature predicts user sign in patterns based on Azure Monitor Logs over the past 7 weeks. This ensures that the VM is ready when the user signs in, while reducing unnecessary resource usage. The system either works alongside the predefined schedule in hybrid mode, where both the predefined and learned schedules are used together, or it primarily relies on its own predictions in automated mode, only falling back to the predefined schedule when it cannot predict user behavior.

Hybrid mode

In hybrid mode, the system pre-stages disks are based on both the predefined schedule and the learned schedule. The disk will remain in the running state during both time frames. This mode combines the benefits of a fixed schedule with adaptive learning to accommodate user behavior that deviates from the standard schedule.

Automated mode:

In automated mode, the system uses only the learned schedule to adjust the disk state. If there is insufficient data from Azure Monitor Logs, the system defaults to the predefined schedule. For new users (less than 1 month), the system initially follows the predefined schedule until enough data is collected for intelligent pre-staging.

Use case examples

The following examples demonstrate different scenarios and the expected behavior based on using the schedule for Wednesdays.

Use Case 1:

Scenario: Over the last 3 weeks, the user logged into the host at least 2 Wednesdays.

Expected Behavior: The system calculates the new schedule by selecting the earliest and latest login times from those 2 or 3 days.

Use Case 2:

Scenario: Over the last 3 weeks, the user signed in only once on Wednesdays.

Expected Behavior: The system extends the data range to 7 weeks and, if it finds at least 2 days of sign ins on Wednesdays, calculates the new schedule using the earliest and latest sign in times from those Wednesdays.

Use Case 3:

Scenario: Over the last 3 weeks, the user signed in only once on Wednesdays, and no additional sign ins were found over the extended 7-week period.

Expected Behavior: The system determines there is insufficient data to calculate an intelligent schedule and defaults to the pre-configured schedule.

Use Case 4:

Scenario: Over the last 3 weeks, there has been no sign in data for Wednesdays.

Expected Behavior: The system checks the past 4 weeks to see if there is another day of the week when the user signed in at least twice (e.g., 2 Tuesdays). If such a day exists, the system assumes Wednesday is a non-working day for the user. Pre-staging will not change disk type for the host this day.

Use Case 5:

Scenario: Over the last 3 weeks, there is no sign in data for Wednesdays, and no other day of the week in the past 4 weeks has 2 days of sign ins.

Expected Behavior: The system determines there is insufficient data to calculate a schedule and defaults to the pre-configured schedule.

Use Case 6:

Scenario: The user signed in multiple times, but only on the last Wednesday, with no additional sign ins recorded in the past 7 weeks.

Expected Behavior: The system aggregates all sign ins and sign outs within a single day as one working session. As a result, the system determines there is insufficient data to calculate an intelligent schedule and defaults to the pre-configured schedule.