Each PVS server independently runs a dynamic rebalance process at regular intervals (~10 minutes)

The process:

• The system enumerates all connected target devices on the server
• Target devices are grouped based on the vDisk used
• Calculation completed to determine whether rebalancing is required based on Fair Load across PVS servers in the site.
• If rebalancing is needed:
  • The system identifies the vDisk with the highest number of targets
  • A subset of those targets are randomly selected and sent for reconnect.
    • The number of targets selected is to achieve Fair Load up to maximum number of targets using that vdisk
    • Only 1 vdisk per cycle is selected and targets sent for rebalance. 
  • These targets reconnect and will be redirected to the least loaded PVS server on reconnection, so they may attach to a different PVS server
• The next pass is then completed a minute later, enumerating all connected targets, grouping them, and calculating if rebalance is required and then sending selected targets for reconnect.
• Next interval in ~60 seconds, or if fairly loaded in ~10 minutes.

Key Design Behavior

• vDisk-driven prioritization
  • Rebalance decisions are based on which vDisk has the highest load, not on overall fairness across workloads.
• No per-workload balancing guarantee
  • Although total device counts may become evenly distributed:
  • Devices using specific vDisks (e.g. single-session vs multi-session) may remain unevenly split
• Reactive—not predictive
  • Auto Rebalance: Responds to current imbalance
  • Auto Rebalance: Does not proactively optimize distribution across all workload types

Auto Rebalance is not done in below scenarios:

• When the PVS server has less than 5 target devices connected.
• If the number of connected targets devices which are booting is greater than or equal to 20% of the total connected devices.
• If the database is offline.

PVS administrators may observe in some scenarios only targets using the same vdisk are always selected for rebalance.

A simple example is if the majority of target devices in the environment are streaming from a single vdisk, targets using that vdisk will be selected and sent for reconnect during Auto Rebalance.

This article explains underlying logic and illustrates associated behavior.