About Power Management and Load Balancing for Farms in Horizon Cloud

This topic describes how Agent Load Index, based on load balancing settings, is used for power management in RDSH farms in Horizon Cloud.

Horizon Cloud agents use five settings ( CPU Usage Threshold, Memory Usage Threshold, Disk Queue Length Threshold, Disk Read Latency Threshold, and Disk Write Latency Threshold) to calculate the Agent Load Index, a value between 0 and 100 that measures each VM's load. For more information about these settings, see Create a Farm.

Important: Because of the key role that Agent Load Index plays in power management, it is essential that you select appropriate values for these settings so you can achieve the desired balance of power consumption and performance in your environment.

Usage of a farm is determined by selecting the higher of the following two percentage values:

Session occupancy - The number of active sessions within a farm divided by the total number of sessions possible on the powered-on VMs in the farm. The number of sessions possible is calculated by multiplying the number of powered-on VMs in the farm by the Sessions per VM value you set for the farm. For more information on the Sessions per VM setting, see Create a Farm,
Average load index - The average Agent Load Index of the powered-on VMs in the farm.

For farm expansion, that value is then compared to the high threshold for the Power Management setting you selected for the farm. For more information about the Power Management setting, see Create a Farm.

In both of the following examples, the Power Management setting for the farm is Optimized Performance. The high threshold for the Optimized Performance setting is 50%, meaning that when the usage reaches 50%, the system powers up one of the unused VMs.

Note: In the examples below, the Max VMs setting for the farm must be greater than 1. Otherwise, the expansion does not occur. For more information about the Max VMs setting, see Create a Farm.

Example - Farm Expansion Due to Session Occupancy Exceeding the High Threshold

In this example, settings are as follows:

Sessions per VM = 20
High threshold for Power Management = 50%

Before Expansion	After Expansion
Powered-on VMs VM 1 Sessions running = 10 Agent Load Index = 25% Usage values Session occupancy = 10 sessions running / (20 Sessions per VM x 1 VM) = 50% Average load index = Agent Load Index 25% / 1 VM = 25% The higher of the two values is 50%, which matches the high threshold for the Optimum Performance setting for Power Management. As a result, the system powers on a second VM.	Powered-on VMs VM 1 Sessions running = 10 Agent Load Index = 25% VM 2 Sessions running = 0 Agent Load Index = 0% Usage values Session occupancy = (Sessions running 10 + 0) / (20 Sessions per VM x 2 VM) = 25% Average load index = (Agent Load Index 25% + 0%) / 2 VMs = 12.5% The higher of the two values is 25%, which is below the high threshold for the Optimum Performance setting for Power Management. As a result, the system takes no action.

Before Expansion

After Expansion

Powered-on VMs

VM 1
- Sessions running = 10
- Agent Load Index = 25%

Usage values

Session occupancy = 10 sessions running / (20 Sessions per VM x 1 VM) = 50%
Average load index = Agent Load Index 25% / 1 VM = 25%

The higher of the two values is 50%, which matches the high threshold for the Optimum Performance setting for Power Management. As a result, the system powers on a second VM.

Powered-on VMs

VM 1
- Sessions running = 10
- Agent Load Index = 25%
VM 2
- Sessions running = 0
- Agent Load Index = 0%

Usage values

Session occupancy = (Sessions running 10 + 0) / (20 Sessions per VM x 2 VM) = 25%
Average load index = (Agent Load Index 25% + 0%) / 2 VMs = 12.5%

The higher of the two values is 25%, which is below the high threshold for the Optimum Performance setting for Power Management. As a result, the system takes no action.

Example - Farm Expansion Due to Average Load Index Exceeding the High Threshold

In this example, settings are as follows:

Sessions per VM = 20
High threshold for Power Management = 50%

Before Expansion	After Expansion
Powered-on VMs VM 1 Sessions running = 5 Agent Load Index = 50% Usage values Session occupancy = 5 sessions running / (20 Sessions per VM x 1 VM) = 25% Average load index = Agent Load Index 50% / 1 VM = 50% The higher of the two values is 50%, which matches the high threshold for the Optimum Performance setting for Power Management. As a result, the system powers on a second VM.	Powered-on VMs VM 1 Sessions running = 5 Agent Load Index = 50% VM 2 Sessions running = 0 Agent Load Index = 0% Usage values Session occupancy = (Sessions running 5 + 0) / (20 Sessions per VM x 2 VMs) = 12.5% Average load index = (Agent Load Index 50% + 0%) / 2 VMs = 25% The higher of the two values is 25%, which is below the high threshold for the Optimum Performance setting for Power Management. As a result, the system takes no action.

Before Expansion

After Expansion

Powered-on VMs

VM 1
- Sessions running = 5
- Agent Load Index = 50%

Usage values

Session occupancy = 5 sessions running / (20 Sessions per VM x 1 VM) = 25%
Average load index = Agent Load Index 50% / 1 VM = 50%

The higher of the two values is 50%, which matches the high threshold for the Optimum Performance setting for Power Management. As a result, the system powers on a second VM.

Powered-on VMs

VM 1
- Sessions running = 5
- Agent Load Index = 50%
VM 2
- Sessions running = 0
- Agent Load Index = 0%

Usage values

Session occupancy = (Sessions running 5 + 0) / (20 Sessions per VM x 2 VMs) = 12.5%
Average load index = (Agent Load Index 50% + 0%) / 2 VMs = 25%

The higher of the two values is 25%, which is below the high threshold for the Optimum Performance setting for Power Management. As a result, the system takes no action.